Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 41 views

    ISLR Chap 4 Shaheryar

    Uploaded by

    Shaheryar Zahur
    This document discusses various classification techniques including logistic regression, linear discriminant analysis, quadratic discriminant analysis, and Naive Bayes classification. It notes that classification involves assigning observations to categories or classes and predicts the probability of an observation belonging to a category. Logistic regression models this probability using a logistic function to produce values between 0 and 1. Multiple logistic regression accounts for association between predictors. Multinomial logistic regression handles variables with more than two classes by selecting a baseline class for comparison.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd

    ISLR Chap 4 Shaheryar

    Uploaded by

    Shaheryar Zahur
    41 views16 pages
    This document discusses various classification techniques including logistic regression, linear discriminant analysis, quadratic discriminant analysis, and Naive Bayes classification. It notes that classification involves assigning observations to categories or classes and predicts the probability of an observation belonging to a category. Logistic regression models this probability using a logistic function to produce values between 0 and 1. Multiple logistic regression accounts for association between predictors. Multinomial logistic regression handles variables with more than two classes by selecting a baseline class for comparison.

    Original Description:

    Introduction to Statistical Learning - Chapter 4 Presentation

    Copyright

    © © All Rights Reserved

    Available Formats

    PPTX, PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    This document discusses various classification techniques including logistic regression, linear discriminant analysis, quadratic discriminant analysis, and Naive Bayes classification. It notes that classification involves assigning observations to categories or classes and predicts the probability of an observation belonging to a category. Logistic regression models this probability using a logistic function to produce values between 0 and 1. Multiple logistic regression accounts for association between predictors. Multinomial logistic regression handles variables with more than two classes by selecting a baseline class for comparison.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    41 views16 pages

    ISLR Chap 4 Shaheryar

    Uploaded by

    Shaheryar Zahur
    This document discusses various classification techniques including logistic regression, linear discriminant analysis, quadratic discriminant analysis, and Naive Bayes classification. It notes that classification involves assigning observations to categories or classes and predicts the probability of an observation belonging to a category. Logistic regression models this probability using a logistic function to produce values between 0 and 1. Multiple logistic regression accounts for association between predictors. Multinomial logistic regression handles variables with more than two classes by selecting a baseline class for comparison.

    Copyright:

    © All Rights Reserved
    Download as PPTX, PDF, TXT or read online from Scribd
    Download as pptx, pdf, or txt
    of 16

    Classification

    By
    Syed Shaheryar Zahur

    Convergent Business Technologies.


    Overview
    • Approach for predicting qualitative responses
    • Involves assigning observation to a category or class
    • Predicts the probability of observation’s belonging to a category

    Convergent Business Technologies.


    Why not Linear Regression?
    1. Cannot accommodate a qualitative response with more than two
    classes
    2. Will not provide meaningful estimates with just two classes

    Convergent Business Technologies.


    Logistic model
    • Models probability that response belongs to a particular category
    • In order to get probabilities within range 0 and 1 a logistic function
    is used

    • For estimation of parameters:


    Maximum likelihood is preferred

    Convergent Business Technologies.


    Multiple Logistic Regressions
    • There may be some association between the different predictors

    • Confounding phenomenon

    Convergent Business Technologies.


    Multinomial Logistic Regression
    • A variable with more than two classes
    • Process
    • Select a baseline
    • Then use model

    • Selection order for baseline is unimportant


    Convergent Business Technologies.
    Generative Models
    • Why need for another method
    I. When substantial difference between classes - parameters for logistic
    regression unstable
    II. If distribution of predictors approximately normal and sample size is
    small – more accurate
    III. Methods can be extended to the case of more than two response
    variables

    Convergent Business Technologies.


    Linear Discriminant Analysis
    • Finds best linear equation that clearly
    separates the data into classes
    • Assumptions
    • Distribution of variables is normal
    • Shared covariance across all classes

    Convergent Business Technologies.


    Linear Discriminant Analysis

    threshold = 0.5 threshold = 0.2

    Convergent Business Technologies.


    Linear Discriminant Analysis
    • ROC – a graph showing performance of classification model at all
    classification thresholds

    Convergent Business Technologies.


    Quadratic Discriminant Analysis
    • Assumptions
    • Distribution of variables is normal
    • Each class has its own covariance matrix
    • More flexible than LDA

    • Used when
    • Variance of classifier is not a concern

    Convergent Business Technologies.


    Naïve Bayes Classifier
    • Assumptions
    • Within class, predictors are independent

    • Introduces some bias but reduces variance

    Convergent Business Technologies.


    Comparison

    Bayes boundary line linear Bayes boundary line Bayes boundary line
    20 observations linear linear
    Uncorrelated random 20 observations 50 observations
    variables with different mean Random variables with Non normal distribution
    in each class different mean in each Correlation of 0.5
    class
    Correlation of -0.5
    Convergent Business Technologies.
    Comparison

    Non-linear decision Non-linear decision Non-linear decision


    boundary boundary boundary
    Correlation of 0.5 Normal distribution Normal distribution
    and -0.5 within Complicated non- Sample size = 6
    classes linear function of
    predictors

    Convergent Business Technologies.


    Poisson Distribution
    • Response can be neither qualitative nor quantitative
    E.g Counts – non-negative integer value
    • Distinction from linear regression model
    • Interpretation
    • Mean-Variance Relationship
    • Non-negative fitted values

    Convergent Business Technologies.


    Thank you.

    ©2019 Convergent Business Technologies

    Convergent Business Technologies.

    You might also like