Used Car Price Prediction

Submitted by:
HINAL SETH
 ACKNOWLEDGMENT
I would like to thank Flip Robo Technologies, for giving me this golden
opportunity to work on this valuable project. I got to learn a lot from this project
about Data Scraping, Data Wrangling, and practical implementations of using
machine learning modules.
I take this opportunity to express my profound gratitude and deep regards to
my mentor Ms. Swati Mahaseth for her exemplary guidance, monitoring and
constant encouragement throughout the course of this assignment. The
blessing, help and guidance given by her time to time shall carry me a long way
in the journey of life on which I am about to embark.

Lastly, I thank almighty, my parents, brother, sister and friends for their constant
encouragement without which this assignment would not be possible.

References:

1. https://www.olx.in/
2. https://www.cardekho.com/
3. https://www.cars24.com/
4. https://autoportal.com/
 INTRODUCTION

• Business Problem Framing

The pre-owned vehicle market is a developing business with a market
esteem that has almost multiplied itself in earlier years. The ascent of
online sites and other instruments like it have made it more
straightforward for the two purchasers and merchants to improve
comprehension of the variables that decide the market worth of a pre-
owned vehicle. In light of a set of variables, Machine Learning calculations
might be used to conjecture the cost of any vehicle. The informational
collection will remember data for an assortment of vehicles. There will be
data with respect to the vehicle's specialized components, for example,
the motor kind, fuel type, total driven kilometers, and that's only the tip
of the iceberg, for each vehicle.

There is no all inclusive instrument for building up the retail cost of utilized
vehicles in light of the fact that unique sites utilize various techniques to
make it. By utilizing measurable models to expect to value, it is
conceivable to acquire a fundamental value gauge without entering every
one of the subtleties into the ideal site. The fundamental motivation
behind this study is to think about the precision of two distinct
expectation models for assessing a pre-owned vehicle's retail cost.
Subsequently, we offer a Machine Learning-based philosophy at
anticipating the costs of secondhand vehicles in light of their attributes.

This philosophy can help purchasers hoping to buy a pre-owned vehicle in

making more informed decisions. Clients can now search for all vehicles
in a district without actual endeavors, whenever and from any area.

With the Coronavirus sway on the lookout, we have seen lot of changes
in the vehicle market. Presently some vehicles are sought after
subsequently making them exorbitant and some are not popular
consequently less expensive. With the adjustment of market due to
Coronavirus 19 effect, people/sellers are facing issues with their past Car
Price valuation AI/Machine Learning models. Along these lines, they are
searching for new AI models from new information. Here we are building
the new car price valuation model.

The primary point of this venture is to create a dataset with the help of
web scraping and anticipate the cost of trade-in vehicle in view of
different elements.

• Conceptual Background of the Domain Problem

The costs of new vehicles in the business is fixed by the producer for
certain extra expenses brought about by the Government as assessments.
Along these lines, clients purchasing another vehicle can be guaranteed
of the money/investment they contribute to be commendable. Be that as
it may, because of the expanded cost of new vehicles and the ineptitude
of clients to purchase new vehicles because of the absence of assets,
utilized vehicles deals are on a worldwide increment. There is a
requirement at a pre-owned vehicle cost expectation framework to
successfully decide the value of the vehicle utilizing an assortment of
highlights. Despite the fact that there are sites that offers this assistance,
their expectation technique may not be awesome. Additionally, various
models and frameworks might contribute on anticipating power for a pre-
owned vehicle's genuine market esteem. It is essential to realize their
genuine market esteem while both trading.

To have the option to anticipate utilized vehicles market worth can help
the two purchasers and merchants. Utilized Vehicle merchants are one
of the greatest objective gathering that can be keen on consequences of
this review. On the off chance that pre-owned vehicle merchants better
get what makes a vehicle attractive, what the significant highlights are
for a pre-owned vehicle, then, at that point, they might think about this
information and proposition a superior assistance.

• Review of Literature
With the recent arrival of internet portals, buyers and sellers may obtain
an appropriate status of the factors that ascertain the market price of a
used automobile. Lasso Regression, Multiple Regression, and Regression
Trees are examples of machine learning algorithms. We will try to develop
a statistical model that can forecast the value of a pre-owned automobile
based on prior customer details and different parameters of the vehicle.
This project aims to compare the efficiency of different models'
predictions to find the appropriate one. On the subject of used
automobile price prediction, several previous studies have been
conducted.

We did a background survey regarding the basic ideas of our project and
used those ideas for the collection of information like the technological
stack, algorithms, and shortcomings of our project which led us to build a
better project.

OLX
The OLX marketplace is a location where you can buy and sell services and
things including electronics, clothing, furniture, household goods,
vehicles, and motorcycles. According to reports, the network had 11
billion page views in 2014, 200 million monthly active users, 25 million
listings, and 8.5 million monthly transactions.

Car Dekho
CarDekho is one of India's most popular automobile websites for new and
secondhand car research. It offers precise automobile prices on the road,
as well as authentic user and expert evaluations. It may also use the
automobile comparison tool to compare different cars. This service also
allows you to connect with local vehicle dealers to find the greatest deals.

AutoPortal
AutoPortal is a digital platform that allows users to study New Cars in India
by looking at prices, specifications, images, mileage, reviews, and
comparisons. Here, you can easily Sell Used Car to verified customers.
One may offer their used automobile for sale with images, models, years
of purchase, and kilometres so that it can be seen by thousands of
potential customers in their city. User reviews and professional
automobile reviews with photographs are available to assist in making a
new car purchase decision.
 Cars24
Cars24 is a website where used car sellers may list their vehicles for sale.
It's an Indian start-up with a simple user interface that asks sellers for
information like automobile model, mileage, registration year, and
vehicle type (petrol, diesel). These enable the online model to estimate
the price by running particular algorithms on provided parameters.

• Motivation for the Problem Undertaken

Objective of the Project:
1. Data Collection : To scrape the data of at least 5000 used cars from
various websites like olx, cardekho, cars24, autoportal, cartrade, etc.
2. Model Building : To build a supervised machine learning model for
forecasting value of a vehicle based on multiple attributes.

Motivation Behind the Project:

There are a few major worldwide multinational participants in the
automobile sector, as well as a number of merchants. By trade,
international companies are mostly manufacturers, although the retail
industry includes both new and used automobile dealers. The used
automobile market has seen a huge increase in value, resulting in a
bigger percentage of the entire market. In India, about 3.4 million
automobiles are sold each year on the secondhand car market.
 Analytical Problem Framing
• Data Sources and their formats
There has been a continuous paradigm of commodity exchanges in
existence for a long time. Previously, these transactions were conducted
through a barter system, which was ultimately converted into a
monetary system. And, as a result of these considerations, any changes
in the pattern of re-selling things were also affected. The resale of an
object can be accomplished in two ways.
The first is offline, while the second is online. In offline transactions,
there is a middleman who is extremely susceptible to corruption and
making excessively lucrative deals. The second alternative is to sell it
online, where there is a platform that allows the user to find out what
price he may earn if he sells it.
We scrape the data for 5000+ cars from websites like olx, cars24,
cardekho and AutoPortal. And save it for Machine Learning Model.
We have extracted attributes like Brand, model name alongwith it’s
manufacturing year; Varient ; Fuel type; number of owners; location;
Date of posting ad online; transmission; driven kilometers and lastly,
price. Price is a int type column and is our target variable. Rest all
attributes are of object datatype.

• Mathematical/ Analytical Modeling of the Problem

We find out that attributes like brand, model, manufacturing year,
varient, total driven kilometers, date of posting ad online and
location have very wide range of variables and it’s not very smart to
study their plots. We do not need to perform bivariate analysis on
these.
• Data Preprocessing Done
We observe that there are 39 missing values in column ‘varient’ and
since it is a column with values in categorical datatype , we will
replace the null values with mode.

• Software Requirements and Tools Used

Python was the most popular technology for implementing machine
learning ideas, owing to the fact that it has a large number of built-in
algorithms in the form of bundled libraries. The following are some of
the most important libraries and tools we used in our project :

1. Numpy:
NumPy is a Python module for array processing. It includes a high-
performance multidimensional array object as well as utilities for
manipulating them. It is the most important Python module for
scientific computing. NumPy may be used as a multi-dimensional
container of general data in addition to its apparent scientific
applications.
Numpy allows any data types to be created, allowing NumPy to
connect with a broad range of databases cleanly and quickly.

2. SciPy:
SciPy is a Python library for scientific and technical computing that is
free and open-source. Optimization, linear algebra, integration,
interpolation, special functions, FFT, signal and image processing,
ODE solvers, and other activities used in research and engineering are
all covered by SciPy modules.
SciPy is based on the NumPy array object, and it's part of the NumPy
stack, which also contains Matplotlib, pandas, and SymPy, as well as a
growing number of scientific computing libraries. Other apps with
comparable users to NumPy include MATLAB, GNU Octave, and
Scilab.
The SciPy stack is occasionally used interchangeably with the NumPy
stack. The SciPy library is now available under the BSD licence, with
 an open community of developers sponsoring and supporting its
development.

3. Scikit-Learn
Scikit-learn offers a standard Python interface for a variety of
supervised and unsupervised learning techniques. It is provided
under several Linux distributions and is licenced under a liberal
simplified BSD licence, promoting academic and commercial use. The
library is being constructed.

4. Jupyter Notebook
Jupyter Notebook is an open-source online software that lets you
create and share documents with live code, equations, visualisations,
and narrative prose. Data cleansing and transformation, numerical
simulation, statistical modelling, data visualisation, machine learning,
and more are all included.
Jupyter Notebook is an open-source online software that lets you
create and share documents with live code, equations, visualisations,
and narrative text. Data cleansing and transformation, numerical
simulation, statistical modelling, data visualisation, machine learning,
and more are all included.
 Model/s Development and Evaluation

• Identification of possible problem-solving

approaches (methods)
We go over the many techniques and datasets that were used to create
this module.
The model will be trained using a dataset comprising over 5000 tuples.
The value of a car is determined by factors such as the number of
kilometres driven, the year of registration, the kind of gasoline used, and
the financial strength of the owner. We created regressor methods and
compared the two on different car models because this is a regression
problem.
Anaconda seeks to address Python's dependency hell, where distinct
projects have various dependency versions, so that project
dependencies do not require separate versions, which might conflict.

• Testing of Identified Approaches (Algorithms)

The models used training and testing datasets are as followed:
1. SVR
2. Linear Regression
3. SGD Regressor
4. KNeighbors Regressor
5. Decision Tree Regressor
6. Random Forest Regressor

• Run and Evaluate selected models

1. SVR:
SVR is based on the same principles as SVM, with a few small
exceptions. It tries to determine the curve given a set of data points.
However, because it is a regression technique, rather of utilising the
 curve as a decision boundary, the curve is used to identify a match
between the vector and the curve's location. Support Vectors aid in
establishing the most accurate match between data points and the
function used to represent them.

2. Linear Regression
Regression is a method for predicting a dependent component
with the help of independent variables.
The method is commonly used to predict and calculate
correlations between independent and dependent variables. The
regression model establishes a linear or exponential connection
between independent and dependent variables.
Linear regression is a type of regression analysis in which the
independent(x) and dependent(y) variables can be constrained
in a linear relationship. The red line in the graph above is known
as the best fit straight line. We want to draw a line that best
predicts the data points given the data points we have. The line
may be represented using the linear equation below.
y = a0 + a1 * x # Linear Equation
3. SGD Regressor
The loss gradient is calculated each sample at a time, and the
model is updated along the way using a decreasing strength
schedule. SGD stands for Stochastic Gradient Descent (aka
learning rate).

The regularizer is a penalty applied to the loss function that

decreases model parameters towards zero using either the
squared euclidean norm L2 or the absolute norm L1 or a mix of
the two (Elastic Net). The update is trimmed to 0.0 whenever the
parameter update passes the 0.0 value due to the regularizer,
allowing for the learning of sparse models and online feature
selection.

4. KNeighbors Regressor
Algorithm Calculating the average of the numerical goal of the K
nearest neighbours is a straightforward implementation of KNN
regression. An inverse distance weighted average of the K
closest neighbours is another method. The distance functions
used in KNN regression are the same as those used in KNN
classification.
5. Decision Tree Regressor
To get from observations about an item (represented in the
branches) to inferences about the item's goal value, decision
tree learning employs a decision tree (as a predictive model)
(represented in the leaves). In statistics, data mining, and
machine learning, it is one of the predictive modelling
methodologies. Classification trees are tree models in which the
goal variable can take a discrete set of values; in these tree
structures, leaves indicate class labels and branches represent
feature combinations that lead to those class labels. Regression
trees are decision trees in which the target variable can take
continuous values (usually real numbers). The objective is to
build a model that predicts the value of a target variable from a
set of input variables.

6. Random Forest Regressor

A regressor with a random forest. A random forest is a meta
estimator that employs averaging to increase predicted accuracy and
control over-fitting by fitting a number of classification decision trees
on various sub-samples of the dataset.
• Visualizations
We have plotted histograms and distribution plot in univariate analysis,
which interpreted that all the columns are equally important but the
columns like brand, variant, location, date and total driven kilometers
have a wide range of data spread hence we will not perform it’s
bivariate analysis.
From bivariate analysis we conclude that, Since Brands, Varients, Driven
Kilometers & Location have a wide range of values in them, we will not
perform bivariate analysis for them as they will not give us any specific
details. Now by plotting graph of Fuel type, Transmission and Owner
against Price, we conclude that Car that uses Diesel, have automatic
Transmission and Has only 1 owner is more likely to have a high price.
The multivariate analysis done by plotting heatmap says that there is no
multicollinearity in the dataset.
 CONCLUSION
• Key Findings and Conclusions of the Study
Car price prediction has picked researchers' interest since it takes a
significant amount of work and expertise on the part of the field expert.
For a dependable and accurate forecast, a large number of unique
attributes are analysed. We employed 6 different machine learning
approaches to develop a model for forecasting the price of used
automobiles. The respective performances of different algorithms were
then compared to discover the one that best suited the existing data set.
The final prediction model was implemented in a python programme.
Furthermore, the model was tested with test data, yielding an accuracy
of 87.76 percent.

• Learning Outcomes of the Study in respect of Data

Science
Using well-known algorithms from Python libraries, we were able to
successfully construct machine learning algorithmic paradigms. On
our dataset, we first do pre-processing and data cleaning. We
trimmed the tuples that contained null values, which accounted for
less than 1% of the total. The findings revealed a positive
relationship between price and kilometres travelled, as well as year
of registration and kilometres travelled.
Negative correlation is related to the notion of inverse proportion,
whereas positive correlation is related to the concept of direct
proportion. The model was trained using over 5000 tuples.

• Limitations of this work and Scope for Future Work

As a part of future work, we aim at the variable choices over the
algorithms that were used in the project. We could only explore two
algorithms whereas many other algorithms which exist and might be
more accurate. More specifications will be added in a system or
providing more accuracy in terms of price in the system i.e.
1) Horsepower
2) Battery power
3) Suspension
4) Cylinder
5) Torque

As we know technologies are improving day by day and there is also

advancement in car technology also, so our next upgrade will include
hybrid cars, electric cars, and Driverless cars.