Prediction of

selling price of
used cars in India
USING REGRESSION AND DECISION
TREE MODELS
MACHINE LEARNING – 21CSC305P

D I V YA M T YA G I – RA2211003030399
A N S H S A C H D E VA – RA2211003030378
P RAT H A M G U L AT I – RA2211003030389
Index
2.Introduction
and 3. Dataset
1. Abstract
Background Description
Research

6.
4. 5.Architecture
Experimental
Methodology of System
Setup

7.Result 8.Conclusion
 The used car market in India
is growing rapidly, with a Accurately predicting the
wide range of factors price is crucial for both
influencing the selling price buyers and sellers.
of vehicles.

In this project, we utilize

Our approach involves
machine learning models,
feature engineering, data
specifically Linear
preprocessing, and model
Regression and Decision
evaluation to determine the
Tree algorithms, to predict
most accurate and efficient
the selling price of used
model for this task.
cars in India.

Abstract
Abstract – Problem
Statement
• Challenge: The used car market in
India is characterized by a wide range
of variables that affect the selling
price. Factors such as brand reputation,
car condition, mileage, and fuel type
contribute to price variability.
Traditional pricing methods often fail to
account for all these variables
effectively.
• Objective: To address this issue, we
developed machine learning models to
predict car prices based on a
comprehensive dataset. Our goal was
to provide an objective, data-driven
approach to pricing used cars.
 Develop Models: Create and train
machine learning models using Linear
Regression and Decision Tree
algorithms to predict the selling price
of used cars.

Evaluate Performance: Assess the

performance of these models using
metrics like Mean Squared Error (MSE)
and R-squared to determine their
Abstract – accuracy and effectiveness.

Objectives of Compare Models: Compare the

results of the Linear Regression and
the Project Decision Tree models to identify which
approach provides better predictions
for used car prices.
Introduction
• The used car market in India is one of
the fastest-growing sectors, driven by
increasing demand for affordable
vehicles. Pricing used cars accurately is
critical to ensure fair deals for both
buyers and sellers. Traditional valuation
methods, often based on experience or
subjective judgment, may not capture
the full range of factors that influence
the price.
• Significance: Machine learning offers a
data-driven approach to this problem,
allowing for more accurate and objective
price predictions. By utilizing models like
Linear Regression and Decision Trees,
we can predict prices based on a range
of factors, making the process more
efficient and reliable.
 Purpose: This presentation explores the machine learning algorithms used
to predict the selling price of used cars in India.

Algorithms Covered :

Linear Regression: A statistical method for predicting a continuous target

variable based on linear relationships.

Decision Tree: A non-linear model that splits data into subsets based on
feature values to make predictions.

Objective: To understand how each algorithm works and how they are
applied to the problem of predicting car prices.

Introduction to Algorithms
Linear Regression Algorithm - Steps Encode Categorical
Handle Missing Features: Convert
Values: Use techniques categorical data (e.g.,
Steps: Data Preparation: like mean imputation or brand, fuel type) into
remove records with numerical formats using
excessive missing data. One-Hot Encoding or
Label Encoding.

Scale Numerical
Objective Function:
Features: Apply scaling Fit the Model: Train the
Minimize the Mean
methods (e.g., Standard Linear Regression model
Model Training: Squared Error (MSE) to
Scaling) to ensure on the prepared training
find the best-fitting line
features contribute data.
through the data.
equally to the model.

Metrics: Assess
performance using R- Predict Prices: Use the
squared (explained trained model to forecast
Model Evaluation: Prediction:
variance), Mean Squared the selling prices of new
Error (MSE), and Mean or unseen cars.
Absolute Error (MAE).
Decision Tree Algorithm - Steps
Handle Missing Encode Categorical
Values: Similar to Linear Features: Convert
Steps: Data Preparation: Regression, preprocess categorical data to
data to manage missing numerical values to be
values. usable by the algorithm.

Build the Tree:

Recursively split the
Splitting Criteria: Use
data based on the most
metrics like Gini Impurity
significant features until
Model Training: or Information Gain to Model Evaluation:
the stopping criterion is
evaluate and choose the
met (e.g., maximum
best splits.
depth, minimum
samples per leaf).

Metrics: Evaluate using Predict Prices: Use the

Accuracy, Precision, constructed tree to
Recall, and F1-Score. predict the selling price
Prediction:
Visualize the tree to by traversing the tree
understand decision based on the feature
paths. values of new cars.
Background Study

Challenges in the Indian

Market: The Indian used car
market is unique due to the
Traditional Methods: Typically,
diversity of car brands, varying
car dealerships and individual
conditions of roads, climate
sellers rely on their experience or
effects, and different buyer
basic online tools to estimate car
preferences. Factors like the
prices. However, these methods
brand, age, mileage, and fuel
often fail to account for complex
type play a significant role in
relationships between various
determining the selling price. In
factors affecting price.
addition, unstructured data and
inconsistent records make pricing
prediction a difficult task.
Problem Statement
• Inconsistency in Price Predictions:
Due to the wide range of factors affecting
the price, predictions can vary
significantly from one evaluator to
another. This inconsistency leads to
mistrust among buyers and sellers.
• Need for Objectivity: A more objective,
data-driven approach can reduce errors in
price estimation, making transactions
smoother and more transparent.
 Objectives:

To analyze the factors influencing the selling price of used

cars in India.

To develop machine learning models that provide accurate

price predictions.

To evaluate and compare the performance of Linear

Regression and Decision Tree models for this task.

Scope :

Objectives The project focuses on cars sold in India, utilizing data from
sources such as online car marketplaces, dealerships, and

and Scope other available datasets.

The models will predict prices based on key features such as
brand, model, year, mileage, fuel type, and ownership history.
The project is limited to Linear Regression and Decision Tree
models, though other models may be considered in future
work.
Dataset Description
- Overview
• The dataset was obtained from
https://www.kaggle.com/ , https://www.cardekho.com/
. It reflects real-world transactions of used cars in
India, providing a comprehensive view of various
factors affecting car prices.
• The dataset includes approximately Mileage records ,
Fuel Type features and Accident History, covering a
broad spectrum of car models, brands, and conditions.
It encompasses data from different regions across
India, allowing for diverse market representation.
• The primary aim of using this dataset is to develop
and evaluate machine learning models that predict
used car prices based on various features.
Dataset Description
Key Features - Part 1
• 1. Brand : Indicates the car's manufacturer (e.g., Maruti,
Hyundai, Honda). The brand significantly impacts the car's
resale value due to factors like brand reputation, customer
loyalty, and perceived quality .
• Example: Maruti Suzuki, being a dominant player in the
Indian market, often results in higher resale values for its
vehicles.
• 2. Model : Specifies the car model (e.g., Swift, i20, City).
Different models within a brand may have varying levels of
demand and resale value .
• Example: The Hyundai i20 is generally more sought after
compared to other models due to its popularity and features.
• 3. Year of Manufacture : The year the car was produced.
Newer cars usually have a higher resale value due to less
depreciation compared to older models .
• Example: A car from 2018 will typically be priced higher
than a car from 2012, given similar conditions and mileage.
Dataset Description
Key Features – Part 2
• 4. Mileage : The total distance traveled by the car in
kilometers. Higher mileage often correlates with increased
wear and tear, impacting the car’s resale price negatively.
• Example: A car with 40,000 km will usually be valued
higher than a similar car with 80,000 km.
• 5. Fuel Type : The type of fuel used by the car (e.g., Petrol,
Diesel, CNG, Electric). Fuel type influences not only the car’s
running cost but also its resale value, with diesel cars
traditionally fetching higher prices .
• Example: Diesel cars may have higher resale values in
regions where diesel is more economical compared to
petrol.
• 6. Transmission : The type of transmission system
(Manual or Automatic). Automatic transmissions are
increasingly preferred in urban settings, potentially affecting
resale values .
• Example: An automatic transmission may add to the car's
resale value in busy metropolitan areas due to ease of
driving.