Turkey Car Market Research

by Alkım Özyüzer, Elif Öztürk, and Görkem Afşin

Supervised by Onur Varol, Ph.D.

Introduction

Welcome to our project repository! In this project, we aimed to analyze the car market in Turkey to understand the factors influencing car prices. By leveraging machine learning, we sought to answer the research question: How do various features such as brand, model year, vehicle type, and color affect car prices in the Turkish market?

---

1. Research and Findings About the Car Market in Turkey

Hypothesis

The price of cars in the Turkish market is influenced by several factors, including brand, model year, vehicle type, and color. Predictive models can be used to estimate car prices based on these factors.

Research Question

How do these features affect car prices in the Turkish market?

---

2. Data Source

We found our data from Kaggle and it is originally from a website called “Sahibinden.com”. The data contains real prices of cars in Turkey market. There are several features in this data such as Date, Brand, Type of car, Model Year and Price. In this part, we also provided some statistics about the data.

---

3. Data Cleaning and Preprocessing

Steps Taken:

1. Handling Missing Values: We used the isnull() function to check for missing values and found none.
2. Removing Duplicates: Identified and removed 1594 duplicate rows.
3. Encoding Categorical Variables: Transformed categorical variables ('Arac Tip Grubu', 'Marka', 'Yakıt Turu', 'Model Yıl') into numerical format using OneHotEncoder.
4. Scaling Numerical Features: Applied StandardScaler to the 'Km' column to standardize its values.

In our data analysis, we began by employing the isnull() library to detect any missing values in our dataset. Fortunately, our investigation revealed that no data values were missing. Additionally, we conducted a thorough examination for duplicates and identified a total of 1594 duplicate columns. To ensure the accuracy of our analysis, we promptly eliminated these duplicates from the dataset. Next, we proceeded to transform the categorical variables into a numerical format suitable for machine learning models. To achieve this, we specifically focused on the columns 'Arac Tip Grubu', 'Marka', 'Yakıt Turu', and 'Model Yıl'. To prepare these columns for encoding, we utilized the OneHotEncoder class, and through the fit() method, the encoder learned the unique categories and their respective numerical representations for each selected column. By performing this encoding, we obtained a transformed dataset that facilitated further analysis and model training. Moreover, we leveraged the ColumnTransformer() function in our code, which allowed us to apply diverse transformations to distinct column types. In particular, we employed the StandardScaler() transformation on the numerical column 'Km'. This transformation standardizes the values, giving them a mean of zero and a standard deviation of one. By scaling the numerical column, we ensured its compatibility with the rest of the dataset and enhanced the accuracy of our subsequent analyses.

---

4. Data Analysis

Overview

We conducted a thorough analysis of the dataset to understand the relationships between various features and car prices.

Top Color Preferences

As the title suggests, this observation provides insightful information about the color preferences of the car market in Turkey, which might affect market demand and car prices.

• Insight: Visualization indicates higher demand for white-colored cars by far.

Mean Price by Model Year for Popular Brands

This examination helps identify trends considering the model year of the car and provides an understanding of varying prices related to model years.

• Insight: Newer cars generally have higher prices. Classic cars from the 1960s also have high demand, indicating their higher value.

Mean Price of Most Expensive Brands

This analysis studies whether there is a correlation between the manufacturer and the prices for the vehicles, potentially leading to answers regarding the effect of brand value on luxury brand pricing.

• Insight: There is a direct correlation between brand reputation and car prices. Luxury brands tend to have higher prices.

Projected Price Change of BMW Vehicles

This visualization aims to observe and diagnose the potential pattern the price observations can exhibit. It attempts to identify the price trend and predict potential price shifts by fitting a polynomial curve with historical data.

• Insight: Provides insights into potential price fluctuations of BMW vehicles over time.

Interpretation

In this part, we investigated the effects of several features on the prices of cars in Turkey:

• Model Year: The visualization confirms that newer cars have higher prices. Classic cars from 1960 also have high demand, indicating their higher value. But in general, we can say that the model year directly influences the car price.
• Mileage: The plot shows that as the kilometers driven increases, car prices decrease. This relationship makes sense since car performance typically decreases as mileage increases. Therefore, we can conclude that mileage is an important factor that affects the price of a car.
• Fuel Type: Electric cars have the highest average price, followed by hybrid cars. The advanced technology required to produce these cars explains the price difference. Thus, fuel type is another factor that influences the car price.
• Color: Mean prices for the top five most frequent colors show that black cars have a higher mean price compared to other colors. However, the mean prices of other colors are relatively close. This suggests that black may be a popular color in the market, but it is not clear if color directly affects the car price as it may be a matter of personal preference.
• Brand: The graph indicates that the brand of a car directly affects its price. Brands with a high reputation and reliability tend to have higher prices. Top brands often provide advanced technology and higher-quality features, justifying their higher car prices.

---

5. Model Prediction Using Linear Regression

Our target variable is the 'Fiyat' column, which represents the price of a vehicle. The goal is to develop a model that can accurately predict the price based on various features. To assess the performance of our model, we split the data into training and test sets. This was accomplished using the train_test_split function from scikit-learn. We randomly assigned 80% of the data to the training set and 20% to the test set. We considered several features to predict the car prices such as the type of the car,brand, leverage(km),the year of the model and the car type group of the model. We did not use color since it has no significant impact on car prices. To deal with categorical columns, we employed the ColumnTransformer and OneHotEncoder from scikit-learn to handle this. Categorical columns ('Arac Tip Grubu', 'Marka', 'Yakıt Turu', 'Model Yıl', 'Arac Tip') were passed to the OneHotEncoder during the preprocessing step. This encoding converted the categorical values into numerical features using binary vectors. To ensure proper scaling of numerical features, we used the StandardScaler from scikit-learn. This was applied to the 'Km' feature, which was included in the dataset. Scaling the features helps prevent any particular feature from dominating the learning process and ensures that all features contribute equally. We employed the LinearRegression algorithm from scikit-learn to create our linear regression model. The r2_score of the model found as %88 which suggests that our model captures a significant portion of the price variation.

Methodology

1. Target Variable: 'Fiyat' (price of a vehicle).
2. Data Split: Used train_test_split to divide the data into training (80%) and test (20%) sets.
3. Features Considered: Type of car, brand, mileage (Km), model year, and car type group.
4. Handling Categorical Data: Employed ColumnTransformer and OneHotEncoder.
5. Scaling: Used StandardScaler for numerical features.
6. Model Used: LinearRegression from scikit-learn.

Results

• R2 Score: 88%, indicating that the model explains a significant portion of the variance in car prices.