Artificial Intelligence & Data Science (Sem VI)

ADC 601 : Data Analytics & Visualization

Module - 2 : Regression Models

(6 Hours)

Instructor : Mrs. Lifna C S

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Topics to be covered

● Introduction to Regression ● Model selection procedures

● Types of Regression ● Leverage;

○ Simple Linear Regression Model ○ influence measures;

○ Multiple Linear Regression model, ○ diagnostics.

○ Polynomial Regression Models; ● Logistic Regression:

○ Interaction models; ○ Logistic Response function and logit,

○ Weighted least squares; ○ Logistic Regression and GLM,

○ ridge regression; ○ Generalized Linear model,

○ loess regression; ○ Predicted values from Logistic Regression,

○ Bootstrapping ○ Interpreting the coefficients and odds ratios,

● Qualitative predictor variables. ● Linear and Logistic Regression:

● Model Evaluation Measures ○ similarities and Differences,

○ Assessing the models.
ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Regression

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Regression examples

Department of Artificial Intelligence & Data Science Courtesy : Types of Regression

ADC601 : Data Analytics & Visualization
 Regression - Types of relationships

1. Deterministic / functional
- the equation exactly describes the relationship between the two variables.
- Eg :

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Regression - Types of relationships

2. Statistical
- Relationship between the variables is not perfect.
- Eg : The response variable y is the mortality due to skin cancer (number of
deaths per 10 million people) and the predictor variable x is the latitude
(degrees North) at the center of each of 48 states in the United States (U.S. Skin
Cancer data)
- Height and weight
- Alcohol consumed and blood alcohol content
- Vital lung capacity and pack-years of smoking
- Driving speed and gas mileage

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Correlation and Regression
● Correlation: is there a relationship between 2 variables?
● Regression: how well a certain independent variable predict dependent
variable?
● CORRELATION != CAUSATION
● In order to infer causality: manipulate independent variable and observe
effect on dependent variable
● Correlation tells you if there is an association between x and y but it
doesn’t describe the relationship or allow you to predict one variable
from the other.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Correlation : Methods for studying Correlation

1. Scatter Diagrams
2. Karl Pearson's Coefficient of Correlation (Covariance Method)
3. Two way Frequency Table (Bivariate Correlation Method)
4. Rank Correlation Method
5. Concurrent Deviation Method

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Correlation - Types of Correlation

● Positive correlation: the two variables change in the same direction.

● Negative correlation: the two variables change in opposite directions.
● No correlation: there is no association or relevant relationship between the two variables.

Scatter Diagrams

● plots the points on a X-Y plane

● Tells the nature of the relationship
between X & Y
● Helps to obtain an approximate
estimation

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Karl Pearson's Coefficient of Correlation (Covariance Method)

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Wikipedia
 Karl Pearson's Coefficient of Correlation (Covariance Method)

1. Calculate Karl Pearson's Coefficient of Correlation between X and Y from

the given table

X 39 65 62 90 82 75 25 98 36 78

Y 47 53 58 86 62 68 60 91 51 84

2. Calculate Karl Pearson's Coefficient of Correlation between Income and

Expenditure of a worker from the given data

Month Jan Feb Mar April May June

Income 451 459 461 461 463 467

Expenditure 433 437 441 451 455 451

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Karl Pearson's Coefficient of Correlation (Covariance Method)

1. Calculate Karl Pearson's Coefficient of Correlation between X and Y from

the given table

X 39 65 62 90 82 75 25 98 36 78

Y 47 53 58 86 62 68 60 91 51 84

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Karl Pearson's Coefficient of Correlation (Covariance Method)

2. Calculate Karl Pearson's Coefficient of Correlation between Income and

Expenditure of a worker from the given data

Month Jan Feb Mar April May June

Income 451 459 461 461 463 467

Expenditure 433 437 441 451 455 451

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Properties of Karl Pearson's Coefficient of Correlation

1. The correlation coefficient between X and Y is same as the correlation coefficient between Y and X
(i.e, rxy = ryx ).
2. The correlation coefficient is free from the units of measurements of X and Y
3. The correlation coefficient is unaffected by change of scale and origin.
Interpretation of Pearson’s Correlation coefficient
The correlation coefficient lies between -1 and +1. i.e. -1 ≤ r ≤ 1
● A positive value of ‘r’ indicates positive correlation.
● A negative value of ‘r’ indicates negative correlation
● If r = +1, then the correlation is perfect positive
● If r = –1, then the correlation is perfect negative.
● If r = 0, then the variables are uncorrelated.
● If r ≥ 0.7 then the correlation will be of higher degree. In interpretation we use the adjective ‘highly’
● If X and Y are independent, then rxy = 0. However the converse need not be true
ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Brain Kart
 Limitations of Karl Pearson's Coefficient of Correlation

1. Outliers (extreme observations) strongly influence the correlation coefficient. If we see outliers in our data,
we should be careful about the conclusions we draw from the value of r. The outliers may be dropped before the
calculation for meaningful conclusion.
2. Correlation does not imply causal relationship. That a change in one variable causes a change in another.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Types of Regression Models

Department of Artificial Intelligence & Data Science Courtesy : Types of Regression

ADC 601 : Data Analytics & Visualization
 Types of Regression Models - based on # independent variables

1. Simple Linear Regression

2. Multiple Linear Regression
3. Polynomial Regression
4. Ridge Regression (L2 regularization) & Lasso Regression (L1 regularization)
5. Logistic Regression
6. Poisson Regression
7. Time Series Regression

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on # independent variables

1. Simple Linear Regression:

● Involves one dependent variable and one independent variable.

● models the relationship between the two variables as a straight line.
● Eg : predicting sales based on advertising expenditure.

2. Multiple Linear Regression:

● Involves one dependent variable and more than one independent variable.
● Models complex relationships by considering multiple factors simultaneously.
● Eg : predicting house prices based on features like square footage, number of bedrooms, and
location.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on # independent variables

3. Polynomial Regression:

● Extends linear regression by introducing polynomial terms of the independent variable.

● captures non-linear relationships.
● represented by a polynomial curve rather than a straight line.
4. Ridge Regression and Lasso Regression:
● Similar to multiple linear regression but includes regularization terms.
● Used to handle multicollinearity and prevent overfitting in multiple linear regression models.
● Ridge regression ⇒ adds a penalty term based on the square of the coefficients,
● Lasso regression ⇒ adds a penalty term based on the absolute values of the coefficients.
● Commonly used in situations where there are highly correlated independent variables.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on # independent variables

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on # independent variables
5. Logistic Regression

● used for binary classification problems.

● models the probability of an event occurring.
● Eg : Predicting the probability of a customer making a purchase (yes/no), fraud detection, etc.
6. Poisson Regression:

● Used when the dependent variable represents counts or frequencies

● Modeling the number of events within a fixed interval,
● Eg: the number of customer arrivals at a service center.
7. Time Series Regression:
● Depends on the specific time series model used
● Applied to data collected over time to model and predict future values based on past observations.
● Eg: Forecasting stock prices, sales, or any variable that exhibits temporal patterns.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on shape of Regression Line

1. Linear Regression
2. Polynomial Regression
3. Exponential Regression
4. Logarithmic Regression
5. Power Regression (Power Law)
6. Sigmoidal (Logistic) Regression
7. Piecewise Regression
8. Quantile Regression

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on shape of Regression Line

1. Linear Regression:

● Regression line : straight line.

● Used when there is a linear relationship between the
independent and dependent variables.

2. Polynomial Regression:

● Regression line : Curve

● (degree of the polynomial determines its shape)
● when the relationship between variables is nonlinear and can
be better represented by a polynomial.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on shape of Regression Line

3. Exponential Regression:

● Regression line : Exponential curve.

● when the dependent variable changes at a constant percentage rate w.r.t independent variable.

4. Logarithmic Regression:

a. Regression line : logarithmic curve.

b. when the rate of change of the dependent variable is proportional to its current value.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on shape of Regression Line

5. Power Regression (Power Law):

○ Regression line : represents a power-law relationship.

○ modeling relationships where one variable's growth is proportional to a power of the other.

6. Sigmoidal (Logistic) Regression:

○ Regression line : S-shaped curve.

○ used for binary classification problems
○ when the dependent variable has a logistic growth pattern.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on shape of Regression Line

7. Piecewise Regression:
○ Regression line : multiple linear segments / piecewise continuous curves (Spline Regression)
○ Involves fitting multiple linear or nonlinear regression models to different segments of the data.
○ Useful when the relationship between variables changes at certain points or intervals.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on shape of Regression Line

8. Quantile Regression:
● Regression Line: different for the quantiles.
● when the variability in the residuals is not constant across all values of the independent variable.
● estimates different quantiles of the dependent variable.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on type of dependent variable

1. Linear Regression
2. Logistic Regression
3. Multinomial Logistic Regression
4. Ordinal Regression
5. Poisson Regression
6. Negative Binomial Regression
7. Survival Analysis (Cox Proportional-Hazards Model)
8. Robust Regression
9. Quantile Regression
10. Ridge Regression and Lasso Regression

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on type of dependent variable
1. Linear Regression:
○ Type of Dependent Variable: Continuous
○ Predicting a continuous outcome variable,
○ Eg: predicting sales, temperature, or height.
2. Logistic Regression:
○ Type of Dependent Variable: Binary (0 or 1)
○ Used for binary classification problems,
○ Eg: predicting whether a customer will buy a product (1) or not (0).
3. Multinomial Logistic Regression:
○ Type of Dependent Variable: Categorical with more than two categories
○ Suitable when the dependent variable has more than two unordered categories,
○ Eg: predicting the type of fruit (apple, orange, banana).

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on type of dependent variable
4. Ordinal Regression / Ordered Logistic Regression:
○ Type of Dependent Variable: Ordered categorical
○ Used when the dependent variable has ordered categories, but the intervals between them are not
assumed to be equal.
○ Eg: predicting the satisfaction level (low, medium, high).
5. Poisson Regression:
○ Type of Dependent Variable: Count data (non-negative integers)
○ Appropriate for modeling count data,
○ Eg: predict the number of customer arrivals, phone calls, or accidents in a day.
6. Negative Binomial Regression:
○ Type of Dependent Variable: Count data with overdispersion
○ Used when the count data exhibit more variability than expected in a Poisson regression, often due to
unobserved heterogeneity.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Types of Regression Models - based on type of dependent variable
7. Survival Analysis (Cox Proportional-Hazards Model):
○ Type of Dependent Variable: Time until an event occurs
○ Applied in medical research, economics, and other fields
○ to model the time until an event (e.g., death, failure, or relapse) occurs.
8. Robust Regression:
○ Type of Dependent Variable: Continuous, resistant to outliers
○ when there are outliers in the data, and traditional linear regression may be sensitive to them.
9. Quantile Regression:
○ Type of Dependent Variable: Conditional quantiles
○ Examining the effect of predictors on different quantiles of the dependent variable,
○ Provides insights into distributional changes.
10. Ridge Regression and Lasso Regression:
○ Type of Dependent Variable: Continuous, with potential multicollinearity
○ Used when dealing with multicollinearity in LR, and to prevent overfitting.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

Given a set of heights (x) and weights

(y) of 10 students, (Student Height
and Weight Dataset).

Which line summarizes the trend

accurately?

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

● A line that fits the data "best" will be one for which the n prediction errors.
● The line which meets "least squares criterion,"

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

● The formulas are determined using

methods of calculus.

● Here, the Best Fitting Line is also known as

● "least squares regression line,"
● "least squares line."
● "estimated regression equation."

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

Trend is positive

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

Trend is negative

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What is the “Best Fitting Line ?”

● a person who is 0 inches tall is predicted to

weigh -267 pounds!
● This happened because of "extrapolation"
● Scope of the model does not include x = 0.
● In general, if the "scope of the model"
includes x = 0, then b0 is the predicted mean
response when x = 0.

● Mean weight to increase by 6.14 pounds for

every additional one-inch increase in height.
● the mean response to increase or decrease
by b1 units for every one-unit increase in x.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression

The equation of the least squares regression line :

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Numeracy
 Simple Linear Regression : Class Assignment - 1

Consider the example below where the mass, y (grams), of a chemical is related to the time,
x (seconds), for which the chemical reaction has been taking place according to the table.
Find the equation of the regression line.

Time Mass
x
y (grams)
(seconds)
5 40
7 120
12 180
16 210
20 240

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Numeracy
 Simple Linear Regression : Class Assignment - 1 Solution

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Numeracy
 Simple Linear Regression : Class Assignment - 2

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Numeracy
 Simple Linear Regression : Class Assignment - 2 Solution

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Numeracy
 Simple Linear Regression : Class Assignment - 3

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Wall Street Mojo
 Simple Linear Regression : Class Assignment - 3 Solution

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Wall Street Mojo
 Simple Linear Regression - What do b0 & b1 estimate?

● Plot illustrates the relationship between

the predictor "high school grade point
average (GPA)" and the response
"college entrance test score."
● Only four groups of students are
considered — those with a GPA of 1,
those with a GPA of 2, ..., and those with
a GPA of 4.
● Data on the entire subpopulation of
students with a GPA of 1, 2, 3 & 4 are
plotted.
● Here,

"population regression line" —

summarizes the trend in the population
between the predictor x and the mean
of the responses

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - What do b0 & b1 estimate?

● Take a sample of three students from each of

the subpopulations — that is, three students
with a GPA of 1, three students with a GPA of
2, ..., and three students with a GPA of 4 —
for a total of 12 students.
● Here,

● Least squares regression line doesn't match

the population regression line perfectly, but
it is a pretty good estimate

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - 4 Conditions

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - 4 Conditions (contd…)

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Simple Linear Regression - What is The Common Error Variance?

● The plot of college entrance test scores for

each subpopulation have equal variance.
● Common variance is denoted as

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Why should we care about ?

● Suppose you have two brands (A and B) of thermometers

- Measure the temperature in Celsius and Fahrenheit using each brand for 10 days.
- Will this thermometer brand (A) yield more precise future predictions …? Or brand (B) ?

● Here, brand B thermometer yield more precise future predictions than the brand A thermometer.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - How to estimate ?

● Sample Variance :
○
○ It is really close to being like an average.
○ Here, the population mean is unknown, estimate with
○ So, divide by n-1, and not n

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - How to estimate ?

● Mean Square Error :

○
○ Subpopulations has distinct x values.
○ Each subpopulation has its own mean

○ Subpopulation mean is estimated using the

estimated regression equation
○ As we use
we lose two degrees of freedom.
Hence, n-2 in the MSE equation.
○ ,
Regression / Residual standard error
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Coefficient of Determination,

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Coefficient of Determination,

Characteristics of

● Association is not causation.

● A large r-squared value, it does not imply that x causes the changes in y.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Caution

1. The coefficient of determination and the correlation coefficient r quantify the strength of a linear
relationship. When = 0% and r = 0, suggests that there is no linear relation between x and y, and
yet a perfect curved (or "curvilinear" relationship) exists.
2. A large value should not be interpreted as meaning that the estimated regression line fits the data
well. Another function might better describe the trend in the data.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Caution
3. The coefficient of determination and the correlation coefficient r can both be greatly affected by
just one data point (or a few data points).
Eg: Relationship between the number of deaths in an earthquake and its magnitude is examined.
- Data on n = 6 earthquakes were recorded, and the fitted line plot on the left was obtained.
- Slope of the line Remove one unusual data point:
- slope of the line changes from +179.5 to -87.1
- Correlation, r = 0.732
- r changes from a +0.732 to -0.960
Magnitude of earthquake increases ⇒ # deaths increases
- changes from 53.5% to 92.1%.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Caution
4. Correlation (or association) does not imply causation.

5. Ecological correlations — correlations that are based on rates or averages — tend to overstate the

strength of an association.

6. A "statistically significant" value does not imply that the slope is meaningfully different from 0.

7. A large value does not necessarily mean that a useful prediction of the response , or

estimation of the mean response , can be made. It is still possible to get prediction intervals or

confidence intervals that are too wide to be useful.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Pearson Correlation Coefficient

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Pearson Correlation Coefficient

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Pearson Correlation Coefficient

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Pearson Correlation Coefficient

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Simple Linear Regression - Pearson Correlation Coefficient

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Test for the Population Correlation Coefficient

● Used to learn of a linear association between two variables, when it isn't obvious
which variable should be regarded as the response.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Steps for Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Multiple Linear Regression

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Multiple Linear Regression - Interpretation

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Multiple Linear Regression - Bivariate Model

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Multiple Linear Regression - Bivariate Model

Usage of Bivariate Model

● To show the correlation between stress and number of working hours along with stress and
days without exercise, as both sets of variables have a positive correlation.
When Bivariate Model cannot be Used
● To show the correlation between depression and hours of sunlight along with depression and
temperature, as hours of sunlight and temperature are not independent of each other.
Minimise the sum of square residuals
● Based on sum of the square residuals for simple regression.
● Used to calculate the equation of the regression line.
● It is very long and has complicated equations, which is why it is usually calculated by a computer.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Multiple Linear Regression - Class Assignment : 1

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Multiple Linear Regression - Class Assignment : 2

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science
 Polynomial Regression

● Form of Linear regression

● To handle the Non-linear relationship between dependent and independent variables

● Add some polynomial terms to linear regression to convert it into Polynomial regression.

● It is modeled as an nth-degree polynomial function.

○ When the polynomial is of degree 2, it is called a quadratic model;

○ when the degree of a polynomial is 3, it is called a cubic model, and so on.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression

● The degree of the polynomial equation depends on the complexity of the

relationship between the independent variable and the dependent variable
● The degree of order which to use is a Hyperparameter, choose it wisely.
○ High degree of polynomial tries to overfit the data,
○ Smaller values of degree, the model tries to underfit,
○ So, find the optimum value of a degree.
● Polynomial Regression models are usually fitted with the method of least squares.
○ minimizes the variance of the coefficients under the Gauss-Markov
Theorem.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression - Evaluation Metrics
● provide information about how well the model fits the data
● used to compare different models or to select the best model for a given problem.
● Some commonly used evaluation metrics for polynomial regression include:
1. Mean Squared Error (MSE):
a. measures the average squared difference between the predicted and actual values.
b. Calculated as the sum of the squared differences divided by the number of observations.
c. The lower the MSE, the better the model performance.
2. Root Mean Squared Error (RMSE):
a. square root of the MSE and provides a measure of the average deviation of the predictions from the actual
values.
b. The lower the RMSE, the better the model performance.
3. R-squared (R2) Score:
● measures the proportion of the variance in the dependent variable that is explained by the independent
variable(s) in the model.
● It ranges from 0 to 1, with higher values indicating better model performance.
4. Adjusted R-squared Score:
● similar to the R-squared score
● takes into account the number of independent variables in the model.
● It is adjusted for degrees of freedom and penalizes the model for including unnecessary independent
variables.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression - Quadratic Model

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression - Solved Problem

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression - Solved Problem

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression - Solved Problem

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Polynomial Regression - Solved Problem

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression

● an extension of the ordinary least squares (OLS) regression method

● allows for the incorporation of weights in the estimation process.
● useful when the homoscedasticity assumption of OLS is violated,
○ (variance of the errors is not constant across all levels of the independent variable(s).
● Basic idea
○ assign different weights to different observations
○ based on the assumption that the variance of the errors is proportional to the inverse of the
weights.
■ Observations with higher weights have lower variances,
■ so, they are given more influence in the estimation process.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression

● Each term in the weighted least squares criterion includes

○ an additional weight, that determines how much each observation in the data set influences the final
parameter estimates
○ it can be used with functions that are either linear or nonlinear in the parameters.

● One of the common assumptions underlying most process modeling methods, including linear and nonlinear
least squares regression,
○ Each data point provides equally precise information about the deterministic part of the total process
variation.
● it is assumed that the standard deviation of the error term is constant over all values of the predictor
or explanatory variables.
○ This assumption, clearly does not hold, even approximately, in every modeling application.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression

● In a weighted fit,
○ less weight is given to the less precise measurements
○ more weight to more precise measurements when estimating the unknown parameters in the model.
● Using weights that are inversely proportional to the variance at each level of the explanatory variables
yields the most precise parameter estimates possible.
● ∑ wi (yi -y’i )^2
○ wi = Weighting factor for the ith calibration standard
■ (w=1 for unweighted least square regression)
○ yi = Observed instrument response for the i th calibration standard
○ y’i = Predicted (or calculated) response for the ith standard
○ ∑ = The sum of all individual values

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression

● Mathematics used in unweighted least squares regression has a tendency

○ to favor numbers of larger value over numbers of smaller value.
○ Thus the regression curves that are generated will tend to fit points that are at the upper calibration
levels better than those points at the lower calibration levels.
● Examples of weighting factors which can place more emphasis on numbers of smaller value are:
○ wi = 1/yi or wi = 1/yi^2
■ wi = weighting factor for the ith calibration standard
● (wi =1 for unweighted least squares regression).
■ yi = observed instrument response (area or height) for the ith calibration standard

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression - Benefits

● Weighted least squares is an efficient method that makes good use of small data sets.
● It also shares the ability to provide different types of easily interpretable statistical
intervals for estimation, prediction, calibration and optimization.
● The main advantage that weighted least squares enjoys over other methods is the ability to
handle regression situations in which the data points are of varying quality.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression - Disadvantages

1. based on the assumption that the weights are known exactly.

○ The exact weights are almost never known in real applications, so estimated weights
must be used instead.
○ The effect of using estimated weights is difficult to assess, but experience indicates
that small variations in the weights due to estimation do not often affect a regression
analysis or its interpretation.
2. When the weights are estimated from small numbers of replicated observations, the results of
an analysis can be very badly and unpredictably affected.
3. use weighted least squares when the weights can be estimated precisely relative to one
another.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression - Disadvantages

● Sensitive to the effects of outliers.

○ If potential outliers are not investigated and dealt with appropriately, they will likely have a
negative impact on the parameter estimation and other aspects of a weighted least
squares analysis.
○ If a weighted least squares regression actually increases the influence of an outlier,
the results of the analysis may be far inferior to an unweighted least squares analysis.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Weighted Least Square (WLS) Regression

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Ridge Regression Model

● also known as Tikhonov regularization or L2 regularization,

● is a linear regression technique that introduces a penalty term to the ordinary least squares (OLS)
objective function.
● Purpose :
○ to address multicollinearity issues
○ prevent overfitting by adding a regularization term that discourages the model from relying too
heavily on any particular predictor variable.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Ridge Regression Model

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Ridge Regression Model

● Shrinkage Parameter (λ):

○ controls the strength of the regularization.
○ A larger λ leads to greater shrinkage of the coefficients, effectively penalizing large coefficients more
severely.
● Bias-Variance Tradeoff:
● Bias to the model in exchange for reduced variance. This can be beneficial in situations where
multicollinearity is present, as it helps stabilize the estimates of the regression coefficients.
● Multicollinearity:
● Ridge Regression is particularly useful when there are high correlations among predictor variables,
which can lead to unstable and unreliable estimates in OLS.
● The regularization term helps to distribute the influence of correlated variables more evenly.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Ridge Regression Model

● Standardization of Variables:
● It is common practice to standardize the predictor variables before applying Ridge Regression.
● Standardization involves subtracting the mean and dividing by the standard deviation for each variable.
○ This ensures that all variables are on a similar scale,
○ regularization term has a consistent impact across predictors.
● No Variable Selection:
● Ridge Regression does not perform variable selection in the same way as methods like LASSO (L1
regularization).
● It tends to shrink all coefficients toward zero, but it rarely sets them exactly to zero.
● Cross-Validation for λ Selection:
● The choice of the shrinkage parameter (λ) is critical.
● Cross-validation techniques, such as k-fold cross-validation, are often employed to select an optimal λ
that balances model complexity and performance.
ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Ridge Regression Model

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Loess Regression Model
● Loess (Locally Weighted Scatterplot Smoothing) regression
● non-parametric regression technique used for estimating relationships between variables.
● useful when dealing with complex and non-linear relationships in the data.
● fits a smooth curve to the data by locally fitting a polynomial regression model to subsets of the
data.
● Key Characteristics:
1. Local Regression:
● by fitting a polynomial model to a subset of the data points within a specified neighborhood (window)
around each point.
● This allows the model to capture local trends in the data.
2. Weighted Regression:
● The fitting process in Loess involves assigning weights to data points based on their proximity to
the point being predicted.
● Points closer to the target point have higher weights, while those farther away have lower weights.
● This weighting emphasizes the influence of nearby points in the local regression
ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Loess Regression Model
3. Polynomial Fitting:
● In each local subset of the data, a polynomial regression model is fitted.
● The degree of the polynomial is typically low (e.g., quadratic or cubic) to avoid overfitting.
4. Smoothing Parameter:
● The degree of smoothing in Loess is controlled by a parameter often denoted as α or τ.
● This parameter determines the size of the local neighborhood and influences the degree of flexibility in the
fitted curve.
● A larger smoothing parameter results in a smoother curve.
5. Residual Weighting:
● beneficial when dealing with heteroscedasticity (varying levels of variability across the data).
6. Adaptive Bandwidth:
● The bandwidth or window size can vary across different regions of the dataset.
● This adaptability helps to capture local features accurately, especially in areas where the relationship
between variables changes.
ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Loess Regression Model
7. Iterative Process:
● The fitting process in Loess is typically iterative.
● After the initial fit, the weights are adjusted based on the residuals, and the process is repeated. This
iteration helps in refining the model and improving the fit.
8. Robustness
● Loess regression is generally robust to outliers since the influence of each point is locally determined.
● Outliers in one region may have minimal impact on the fit in other regions.

ADC601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Analytics Vidya
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

Note :
Residuals are assumed to be normally distributed with a mean near zero and a constant
variance.
Intercept
● estimated income of $7,263 for a newborn female with no education.
● It is important to note that the available dataset does not include such a person.
● The minimum age and education in the dataset are 18 and 10 years, respectively

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

Coefficient for Age

● is approximately equal to one.
● For every one unit increase in a person’s age, the person’s income is expected to increase by $995.
Coefficient for Education:
● for every unit increase in a person’s years of education, the person’s income is expected to increase by
about $1,758.
Coefficient for Gender
● When Gender is equal to zero, the Gender coefficient contributes nothing to the estimate of the expected
income.
● When Gender is equal to one, the expected Income is decreased by about $934.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

Note:
● Coefficient values are only estimates based on the observed incomes in the sample, there is some
uncertainty or sampling error for the coefficient estimates.
● Std. Error
○ provides the sampling error associated with each coefficient
○ used to perform a hypothesis test, using the t-distribution, to determine if each coefficient is statistically
different from zero.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

● If a coefficient is not statistically different from zero, the coefficient and the associated variable in the
model should be excluded from the model.
● In this example, the associated hypothesis tests’ p-values, Pr(>|t|), are very small for the Intercept, Age, and
Education parameters

● For small p-values, as is the case for the Intercept, Age, and Education parameters, the null hypothesis
would be rejected.
● For the Gender parameter, the corresponding p-value is fairly large at 0.13.
● In other words, at a 90% confidence level, the null hypothesis would not be rejected.
● So, dropping the variable Gender from the linear regression model should be considered

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

● Residual standard error : is the standard deviation of the observed residuals.

○ This value, along with the associated degrees of freedom, can be used to examine the variance of the
assumed normally distributed error terms.
● R-squared (R 2 )
○ measures the variation in the data that is explained by the regression model.
○ vary from 0 to 1,
○ closer to 1 : the model is better at explaining the data than values closer to 0.
○ exactly 1 : the model explains perfectly the observed data (all the residuals are equal to 0).
○ R squared can be increased by adding more variables to the model.
○ However, just adding more variables results in overfitting.
● F-statistic : provides a method for testing the entire regression model.
○ t-tests, individual tests were conducted to determine the statistical significance of each parameter.
○ F-statistic and corresponding p-value enable the analyst to test the following hypotheses:

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Hypothesis Testing for

● F-statistic : provides a method for testing the entire regression model.

○ t-tests, individual tests were conducted to determine the statistical significance of each parameter.
○ F-statistic and corresponding p-value enable the analyst to test the following hypotheses:

In this example, the p-value of 2.2e – 16 is small, which indicates that the null hypothesis should be
rejected.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Interaction Models

● Examine the combined effects of two or more independent variables on the

dependent variable.
● provide insights into how the relationship between one independent variable and the
dependent variable depends on the level or presence of another variable.
● Interaction terms are created by multiplying the values of two or more variables.
● Common types of Interaction Models
1. Two-Way Interaction:
2. Three-Way Interaction:
3. Moderation (Interaction with a Moderator):
4. Mediation (Interaction with a Mediator):
5. Cross-Product Terms:
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Interaction Models - Types
1. Two-Way Interaction:
● Involves the interaction between two independent variables.
● includes the main effects of both variables as well as their interaction term.
● Y=β0 +β1 X1 +β2 X2 +β3 (X1 ×X2)+ϵ
2. Three-Way Interaction:
● Involves the interaction between three independent variables.
● The model includes the main effects of all three variables as well as their pairwise and triple
interaction terms.
● Y=β0 +β1 X1 +β2 X2 +β3 X3 +β4 (X1 ×X2 )+β5 (X1 ×X3)+β6 (X2 ×X3 )+β7(X1 ×X2×X3 )+ϵ
3. Moderation (Interaction with a Moderator):
● Examines whether the effect of one independent variable on the dependent variable is moderated by
the presence or level of another variable.
● Y=β0 +β1 X1 +β2 X2 +β3 (X1×X2)+ϵ, Where, X2 moderates the relationship between X1 and Y.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Interaction Models - Types

4. Mediation (Interaction with a Mediator):

● Investigates whether the effect of one independent variable on the dependent variable is mediated by
the presence or level of another variable.
● Y=β0 +β1 X1 +β2 X2 +β3 (X1 ×X2)+ϵ, where X2 mediates the relationship between X1 and Y.
5. Cross-Product Terms:
● More general term for interaction terms
● involve multiplication of various independent variables.
● Y=β0 +β1 X1 +β2 X2 +β3 (X1 ×X2)+β4 (X1 ×X3)+ϵ
Interaction models
- Helps to explore how the effects of one variable may change depending on the values of another variable.
- involves examining the coefficients of the interaction terms and understanding how they influence the overall
relationship between variables.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Qualitative Predictor variables
● also known as categorical variables,
● pose some unique challenges compared to continuous variables.
● can be either nominal or ordinal,
● their inclusion in regression models requires special considerations.
● Some techniques used with qualitative predictor variables:
1. Dummy Coding (Indicator Variables):
● common technique for representing categorical variables in regression models.
● For a categorical variable with k levels, k−1 dummy variables are created.
● Each dummy variable takes the value 0 or 1, ⇒ absence or presence of a particular level of the
categorical variable.
● Example: For a variable "Color" with three levels (Red, Green, Blue), you might create two dummy
variables: D1 for Green and D2 for Blue, with Red as the reference level.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Qualitative Predictor variables
2. Effect Coding:
● Similar to dummy coding,
● reference category is assigned a value of -1, and the other categories are coded as 0 or 1.
● useful when you are interested in the overall average effect of the categorical variable.
3. Contrast Coding:
○ Involves creating contrasts that represent specific comparisons of interest among the levels of the
categorical variable.
○ Popular contrast codings include treatment (dummy) coding and Helmert coding.
4. Interaction with Dummy Variables:
○ Used when both continuous and categorical predictors,
○ to capture potential differential effects.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model evaluation Measures
1. Mean Absolute Error (MAE):

○ Represents the average absolute differences between the observed and predicted values.
2. Mean Squared Error (MSE):

○ Measures the average squared differences between observed and predicted values.
3. Root Mean Squared Error (RMSE):

○ Square root of the mean squared error, providing a more interpretable scale.
4. R-squared (R2):

○ Represents the proportion of the variance in the dependent variable that is predictable from the
independent variables.
○ Ranges from 0 to 1, where 1 indicates a perfect fit.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model evaluation Measures

5. Adjusted R-squared:

○ Adjusts R-squared for the number of predictors in the model, providing a more realistic measure.
○ Penalizes the addition of irrelevant variables that do not improve the model significantly.
6. Mean Absolute Percentage Error (MAPE):

○ Represents the average percentage difference between observed and predicted values.
○ Useful for expressing errors as a percentage of the observed values.
7. Mean Bias Deviation (MBD):

○ Represents the average difference between predicted and actual values.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model Selection Procedures
● choosing the most appropriate model from a set of candidate models.
● find a model that balances goodness of fit with simplicity to avoid overfitting.
1. Stepwise Regression
2. Subset Selection:
3. Regularization Techniques
4. Information Criteria
5. Cross-Validation
6. Bootstrap Resampling
7. Model Comparison
8. Domain Knowledge
9. Model Diagnostics

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model Selection Procedures

1. Stepwise Regression:
● Forward Selection:
■ Starts with an empty model
■ Adds predictors one at a time,
■ selecting the one that most improves the model fit at each step.
● Backward Elimination:
■ Starts with all predictors in the model
■ Removes the one that contributes least to the model fit at each step.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model Selection Procedures

2. Subset Selection
○ Best Subset Selection:
■ Fits all possible combinations of predictors
■ selects the model with the best fit based on a criterion such as the Akaike Information Criterion
(AIC) or the Bayesian Information Criterion (BIC).
■ balance the goodness of fit and the number of parameters in the model.
■ Model with the lowest AIC or BIC is often selected.
■ Formula :
● L is the likelihood of the model.
● k is the number of parameters in the model.
● n is the number of observations in the dataset.
○ Recursive Feature Elimination (RFE):
■ Iteratively removes the least important variable until the desired number of features is reached.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model Selection Procedures
3. Regularization Techniques:
○ Ridge Regression:
■ Introduces a regularization term to the least squares equation,
■ preventing overfitting by penalizing large coefficients.
○ Lasso Regression:
■ Similar to ridge regression
■ but uses the absolute values of coefficients, promoting sparsity and variable selection.
○ Elastic Net Regression:
■ A combination of ridge and lasso regularization, balancing their strengths.
4. Information Criteria:
○ Akaike Information Criterion (AIC):
■ Penalizes models for complexity,
■ favoring simpler models that explain the data well.
○ Bayesian Information Criterion (BIC):
■ Similar to AIC
■ But places a stronger penalty on model complexity.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model Selection Procedures

5. Cross-Validation:
○ k-Fold Cross-Validation:
■ Divide the data into k folds,
■ train the model on k-1 folds,
■ Validate on the remaining fold.
■ Repeat this process k times, rotating the validation set.
○ Leave-One-Out Cross-Validation (LOOCV):
■ Special case of k-fold where k equals the number of observations.
■ Each observation serves as a validation set in turn.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Model Selection Procedures
6. Bootstrap Resampling:
○ Bootstrap Aggregating (Bagging):
■ Create multiple bootstrap samples from the dataset,
■ train models on each sample,
■ average their predictions to reduce variance.
○ Bootstrap Confidence Intervals:
■ Assess the stability and reliability of regression coefficients using bootstrap resampling.
7. Model Comparison:
○ Compare different regression models using statistical tests or information criteria to determine the
most suitable model for your data.
8. Domain Knowledge:
○ Consider the theoretical aspects of the problem and domain knowledge to guide variable selection and
model specification.
9. Model Diagnostics:
○ Use residual analysis, leverage plots, and other diagnostic tools to identify potential issues with the
chosen model.
ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Leverage in Regression
● influence that a single data point can have on the overall fit of a regression model.
● It is a measure of how much a particular observation can affect the estimated regression coefficients.
● important in multiple linear regression, where there are more than one independent variable.
● measure of how far an independent variable value is from the mean of the independent variables.
● Leverage for an observation i in a dataset with n observations is given by Hat Matrix
○ Also known as Projection Matrix

○ defined in terms of the data matrix X:

■ p is the number of coefficients,

■ n is the number of observations
(rows of X) in the regression model.
○ HatMatrix is an n-by-n matrix in the
Diagnostics table.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Leverage in Regression
● Influence and Outliers:
○ Observations with high leverage can have a strong influence on the estimated coefficients.
○ If a data point has high leverage and an extreme value in the dependent variable,
■ it can significantly impact the regression model,
■ potentially leading to outliers or influential points.
● High Leverage Points:
○ Points with high leverage typically have extreme values in one or more independent variables.
○ These points have the potential to disproportionately influence the regression model,
■ especially if they deviate from the overall pattern of the data.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Leverage in Regression
● Diagnostic plots,
○ such as leverage-residual plots,
○ Helps to identify observations with high
leverage.
○ Observations with both high leverage and
large residuals may have a substantial
impact on the model.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Residual Leverage Plot (Regression Diagnostic)
Regression analysis requires some assumptions to be followed by the dataset.
● Observations are independent of each other. It should be correlated to another observation.
● Data is normally distributed.
● The relationship b/w the independent variable and the mean of the dependent variable is linear.
● The data is in homoscedasticity, (variance of the residual is the same for each value of the dependent
variable.)
To perform a good linear regression analysis, check whether these assumptions are violated:
● If the data contain non-linear trends then it will not be properly fitted by linear regression resulting in a high
residual or error rate.
● To check for the normality in the dataset, draw a Q-Q plot on the data.
● The presence of correlation between observations is known as autocorrelation. autocorrelation plot.
● The presence of homoscedasticity (Scale Location plot, the Residual vs Legacy plot.)

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Residual Leverage Plot (Regression Diagnostic)
1. Residual vs fitted plot:
○ This plot is used to check for linearity and homoscedasticity,
○ if there is a linear relationship then it should have a horizontal line with much deviation.
○ If the model meets the condition for homoscedasticity, the graph should be equally
spread around the y=0 line.
● Q-Q plot:
○ This plot is used to check for the normality of the dataset,
○ if there is normality that exists in the dataset then, the scatter points will be distributed
along the 45 degrees dashed line.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Residual Leverage Plot (Regression Diagnostic)

3. Scale-Location plot:
○ It is a plot of square rooted standardized value vs predicted value.
○ This plot is used for checking the homoscedasticity of residuals.
○ Equally spread residuals across the horizontal line indicate the homoscedasticity of residuals.
4. Residual vs Leverage plot:
○ plot between standardized residuals and leverage points of the points.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression

● Statistical method used for binary classification,

predicting the probability that an instance belongs to a
particular class.
● commonly used for classification problems rather than
regression problems.
● Sigmoid Function (Logistic Function):
○ Is used to model the probability that a given
input belongs to a particular class.
○ y is expressed as a linear function of the input
variables

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression
● Probability Prediction:
○ The output of the sigmoid function represents the probability that the instance belongs to the positive
class (class 1): P(y=1)=σ(z)
○ The probability of belonging to the negative class (class 0) is then 1−P(y=1).
○ based on the input variables , the probability of an event

● Basic difference with Linear Regression

○ the values of y are not directly observed.
○ Only the value of in terms of success or failure (typically expressed as 1 or 0, respectively) is
observed.
● Log odds Ratio
○ Using p, f(y) can be rewritten as :

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression

● Maximum Likelihood Estimation (MLE)

○ used to estimate the model parameters, that maximize the chances of observing the given dataset.
● Decision Boundary:
○ A threshold (usually 0.5) is set, and if the predicted probability is above this threshold, the instance is
classified as belonging to the positive class; otherwise, it is classified as belonging to the negative
class.
● Training:
○ The model is trained by adjusting the weights (b0, b1,…,bn) using an optimization algorithm such as
gradient descent.
○ The objective is to minimize a cost function that penalizes the model for making incorrect predictions.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression Applications

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression Adavantages

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression - Problem

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression - Problem

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression - Problem

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression - Problem

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression - Problem

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Generalized Linear Model
● statistical modeling framework that extends the classical linear regression model
● to handle a broader range of data types and distributions.
● assumes normally distributed errors and continuous response variables,
● accommodating various types of response variables and distributional assumptions.
● Key components of a Generalized Linear Model include:
○ Random Component:
■ This part of the model specifies the distributional family of the response variable, which can be chosen
based on the nature of the data. Examples of distribution families include Gaussian (for continuous
data), Binomial (for binary data), Poisson (for count data), and Gamma (for positively skewed
continuous data).
○ Systematic Component:
■ describes how the linear predictor is related to the predictors.
■ It includes a linear combination of the predictor variables, each multiplied by a regression coefficient.
○ Link Function:
■ connects the mean of the distribution (specified by the random component) to the linear predictor in the
systematic component.
■ The choice of link function depends on the distributional family and the characteristics of the data.
Common link functions include the identity, logit, log, and inverse.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Generalized Linear Model

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Logistic Regression Vs Generalized Linear Model
Feature Logistic Regression Generalized Linear Models (GLM)

Objective Binary classification Generalized framework for regression models

Type of Model Specific case of GLM for binary outcomes General framework that includes LR as a special case

Dependent Variable Binary (0 or 1) Can handle various types (e.g., continuous, count)

Can use different link functions based on the distribution

Link Function Logit function (sigmoid)
(e.g., log, identity)

Distribution Binomial Various distributions (e.g., Gaussian, Poisson)

Assumes a binomial distribution of the

Assumption More flexible in terms of distribution assumptions
response variable

Regression problems with different types of response

Use Cases Binary classification problems
variables

Interpretability Coefficients represent log-odds Interpretation depends on the chosen link function

Examples Predicting whether an email is spam or not Predicting house prices, count of events, etc.

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Linear Regression Vs Logistic Regression

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Linear Regression Vs Logistic Regression

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501
 Linear Regression Vs Logistic Regression

ADC 601 : Data Analytics & Visualization Department of Artificial Intelligence & Data Science Courtesy : Stat 501