Car Price

Prediction
 Introduction
● Predicting the price of a used cars has been studied extensively in various
researches.

● Car price prediction is a somehow an interesting and popular problem.

● As per information from the Agency for Statistics of BiH, the percentage
of personal car usage is increased by 2.7% since 2013 and it is likely that
this trend will continue, and the number of cars will increase day by day.

● Accurate car price prediction involves expert knowledge, because price

usually depends on many distinctive features and factors.

● Typically, the most significant ones are present price , brand and model,
age, mileage etc. The fuel type used in the car as well as fuel consumption
per mile highly affect the price of a car due to a frequent changes in the
price of a fuel. Different features like exterior color, type of transmission,
safety, air condition, etc. will also influence the car price.
 BACKGROUND STUDY
● Wu et al. conducted a car price prediction study, by using a neuro-fuzzy
knowledge-based system. Their prediction model produced similar results as
the simple regression model.

● Listian discussed, in her paper, that a regression model that was built using
Support Vector Machines (SVM) can predict the price of a car that has been
leased with better precision than multivariate regression.

● Another approach was given by Richardson in his thesis work . Richardson

applied multiple regression analysis and demonstrated that hybrid cars retain
their value for a longer time than traditional cars.

● Pudaruth applied various machine learning algorithms, namely: k-nearest

neighbors, multiple linear regression analysis, decision trees and naïve bayes
for car price prediction in Mauritius.

● Noor and Jan built a model for car price prediction by using multiple linear
regression.
 PROBLEM STATEMENT
A model to predict the price of a used car should be
developed in order to assess its value based on a variety
of characteristics. Several factors affect the price of a
used car, such as company, model, year, transmission,
distance driven, fuel type, seller type, and owner type.
As a result, it is crucial to know the car's actual market
value before purchasing or selling it.
 CHALLENGES
● Finding the best regression algorithm, among Linear Regression,
Lasso Regression, Random Forest Regression, Ridge regression, etc.,
for our problem was a challenge.

● We had to learn about random forest regression since we weren't

familiar with it.

● Reconstructing the given dataset. Changing the categorical values

into numerical form. And removing the unnecessary features.

● The choice between randomizedsearchCV and gridsearchCV for finding

hyperparameters of random forest regression presented a challenge.

● We had a challenge in choosing dominant features for which we used

heat maps using pearson coefficient correlation and extra tree
regressor and found the feature importances.

● We developed a webpage to demonstrate the workings of our model

for which we had to learn HTML and CSS.
 Solution Approach
● Step 1: "Car_Data.csv" dataset was taken from kaggle.com and was
reconfigured to reflect the important features.

Data frames in which we loaded the dataset now include the Selling price ,
Present price, Kms driven, owner, age, Fuel Type, Seller type, Transmission
type.

● Step 2: Categorical features were then converted to numerical values.

We remove multiple columns of the dataset using the get_dummies function

as some columns contain the same information because the original column
could assume a binary value, The "CNG" fuel type, the "Automatic"
transmission type, and the "Dealer" seller type were removed.

● Step 3: We created a heat map using Pearson correlation to illustrate how the
features are related.
 Solution Approach
● Step 4: Using ExtraTreesRegressor the feature importances were
obtained. And Present_Price turned out to be the feature with most
importance.

● We used Random Forest Regression to solve the problem and created

a model based on this.

● Step 5: By using randomizedsearchCV, we performed hyper parameter

tuning

(n_estimators, max_features, max_depth, min_samples_split,

min_samples_leaves).

Due to its speed, RandomizedSearchCV is better than GridSearchCV.

● Step 6: The data has been split into training data (80%) and test data
(20%).
 Solution APPROACH
● Step 7: After getting the best parameters from the
RandomizedSearchCV we train the model using the Random Forest
Regression .

In the random forest regressor, the decision tree will scale the input,
so we do not have to scale the values.

● The graph of predictions is plotted using displot/ distplot in which

we observe that it resembles a normal distribution with mean 0.

● Step 8 : The file is put in a pickle file. Also the performances of the
model are computed.

● Step 9 : The model is demonstrated by entering features in an HTML

file and predicting the price by using a Python file from the set of
values given as features.

Lastly, the final output is displayed on the HTML page.

 DataSET
The data we have used in this project was downloaded from Kaggle. It was uploaded from
Cardekho.com . The dataset consists of 301 rows and 9 columns with no null values. Column data
consist of independent Features. The independent features contain both categorical and numeric
values.
 Dataset MANIPULATION
● From the original dataset we searched for null values using “isnull()”.

● Removed the “car_name” column from the dataset. We added a new

feature “Age” by subtracting current year and purchase year. Then we
dropped the “Year” column.

● As the dataset has numerical values and categorical values , We

remove multiple columns of the dataset using the get_dummies
function as some columns contain the same information because the
original column could assume a binary value,

● The "CNG" fuel type, the "Automatic" transmission type, and the
"Dealer" seller type were removed from the unique values.
 FINAL DATASET
● Using the Extra Trees Regressor the feature
importances are found.

● The Present Price is the feature with most

importance

● And also we computed the correlation of all the

features , we used “pearson” method.

● And from that the we found that Selling price

and the present price are much correlated.

● And a heatmap is used to plot the correlation

between 2 features.
GITHUB REPO
Link:
iyashk/Car-Price-Prediction(github.com)
 OBSERVATION
● After the model is trained , we plotted the difference between test and train data along with the predictions made
using displot . This is shown in normal distribution with mean 0.

● This tells us that the error is not much. So the predicted value might actually be accurate.

● Also when we use the scatter plot the test points and the predicted points are along the line y=x , which means that
the predicted values are equal to the test values using the model.
 OBSERVATION
● The performance measures are found using the Root Mean Square
Error , R2 Score , Mean Square Error.
● Also accuracy of the model is calculated from the sklearn’s built in
functions. As this is only done for classification we made our
predicted output using a cutoff. If the difference between the test
and predicted value is greater than cutoff (MSE) then we classify as
wrongly predicted else correct prediction.
● Accuracy of the model calculated from the “accuracy_score” gave us
a score of 93.4 percentage.
● Also calculating the Root Mean Square Error for the test and
predicted test values is 1.36.
● Calculating the Mean Square Error for the test and predicted test
values we get 1.97.

accuracy of the model : 0.9344262295081968

 RESULTS
The results are demonstrated in html page. The
features are entered and the selling price is
calculated.

FEATURES
Year : 2007
Present price : 3.5 lakhs
Kms driven : 50000
Owner : 0
Fuel Type : Petrol
Seller Type : Dealer
Transmission type : Manual

PREDICTED PRICE
The price predicted is 2.81 lakhs
 RESULTS
The results are demonstrated in html page. The
features are entered and the selling price is
calculated.

FEATURES
Year : 2015
Present price : 59 lakhs
Kms driven : 5000
Owner : 1
Fuel Type : Diesel
Seller Type : Individual
Transmission type : Automatic

PREDICTED PRICE
The price predicted is 22.13 lakhs
References [1] Gegic, E.; Isakovic, B.; Keco, D.; Masetic, Z.; Kevric, J. Car price
prediction using machine learning techniques. TEM J. 2019, 8, 113.

[2]USA_CAR_SELLING_PRICE/Second-checkpoint.ipynb at master ·
harsh0703-harsh/USA_CAR_SELLING_PRICE (github.com)

[3]
Capstone_Project_Machine_Learning/Capstone_Project_Report_with_
Python.pdf at master ·
EnesGokceDS/Capstone_Project_Machine_Learning (github.com)

[4] sklearn.ensemble.ExtraTreesRegressor — scikit-learn 1.0.2

documentation

[5] sklearn.ensemble.RandomForestRegressor — scikit-learn 1.0.2

documentation

[6] sklearn.model_selection.RandomizedSearchCV — scikit-learn 1.0.2

documentation

[7] used-car-price-prediction/ml-models.ipynb at master ·

abhashpanwar/used-car-price-prediction (github.com)

[8] Predicting Used Car Prices.pdf (stanford.edu)

 THANK YOU!
“Predicting the future isn’t magic
It’s Artificial Intelligence.”
~Dave Waters
CREDITS: This presentation template was
created by Slidesgo, including icons by
Flaticon, and infographics & images by
Freepik and illustrations by Stories