car-price-regression

🚗 Car Price Regression & Classification Project

This repository contains my full machine learning project, built as part of Assignment #2 in the Introduction to Data Science course at Reichman University. The goal of the project was to analyze a large cars dataset, clean and enrich the data, visualize insights, engineer features, and ultimately train two predictive models:

• Regression model — Predicting car prices
• Classification model — Categorizing cars into Low / Mid / High price ranges

Both final models are exported as .pkl files and available here in the repository.

---

📊 1. Dataset Overview

The dataset includes almost 19,000 cars, with features such as:

• Manufacturer & Model
• Production Year
• Mileage
• Engine Volume
• Gearbox Type
• Fuel Type
• Drive Wheels
• Color
• Airbags
• and more.

Before modeling, I performed cleaning, handled outliers, and created new engineered features.

---

🔍 2. Exploratory Data Analysis (EDA)

Key steps:

• Visualizing the raw price distribution
• Applying log-transform to stabilize variance
• Identifying outliers using quantiles
• Removing the top/bottom 1% noisy records
• Plotting cleaned price distribution

This resulted in a cleaner dataset with 18,864 rows, much more suitable for modeling.

---

⚙️ 3. Feature Engineering

I engineered several new features and cleaned others:

• Car_age — calculated from production year
• Engine_volume_clean — cleaned numeric version of engine volume
• Doors_clean — fixed inconsistencies
• Removed unnecessary columns (ID, Levy, Doors, raw Engine Volume)
• One-Hot Encoding for all categorical variables
• Scaling numeric features

🌀 Clustering Feature

I also added a brand-new feature using KMeans clustering (Cluster). This helped group vehicles into 3 natural segments and improved model performance.

---

📈 4. Regression Models

I trained 3 regression models:

1. Linear Regression (baseline + improved)
2. Decision Tree
3. Random Forest (winner)

🏆 Winning Model

The best model was Random Forest Regressor, achieving strong results (lowest MAE/RMSE and highest R²). The final model is exported here as:

car_price_random_forest.pkl

---

🔄 5. Classification

I converted the price variable into 3 balanced classes using qcut:

• 0 = Low
• 1 = Mid
• 2 = High

Then I trained:

• Logistic Regression (🔥 Winner — ~76% accuracy)
• KNN
• Random Forest Classifier

The Logistic Regression model gave the best results and is exported as:

car_classification_logistic.pkl

---

🤖 6. Files Included in This Repository

---

🎬 7. Video Presentation

A video walkthrough of the entire project, including:

• Dataset overview
• EDA
• Feature engineering
• Clustering
• Model training
• HuggingFace upload

Will be attached here.

---

✨ 8. Summary

This project gave me practical experience with:

• Data cleaning
• EDA
• Feature engineering
• Clustering
• Regression & classification models
• Exporting models
• Managing a HuggingFace repository

All work is included in this repository and submitted as required.

---

Thank you for reviewing! Yotam Gil https://colab.research.google.com/drive/1XQCA3yYdm1k0bnV02Gz1eC3uFdy_N0p2#scrollTo=LcN1CpJ0h_tt