Skip to content

Latest commit

 

History

History
84 lines (53 loc) · 3.05 KB

README.md

File metadata and controls

84 lines (53 loc) · 3.05 KB
Raw

Car Price Prediction using Machine Learning

This project applies multiple machine learning models to predict car prices based on numerical, categorical, and image data from the DVM-Car dataset. The goal is to compare the performance of various models, including both classical machine learning algorithms and deep learning models, in accurately predicting car prices.

Project Overview

The project involves predicting car prices using several machine learning models, focusing on both classical tabular data (numerical and categorical) and image data. The models implemented include:

  • K-Nearest Neighbors (KNN)
  • Extreme Gradient Boosting (XGBoost)
  • Multilayer Perceptron (MLP)
  • Convolutional Neural Network (CNN)

Roadmap of the project (see pdf : https://github.com/Olivierjaylet/Car-Price-Forecast/blob/1e337ce00b09a74b5e77ff2d0cf8b58183eb6844/Car_Price_Prediction.pdf)

  1. Data Description
  2. Feature engineering
  3. Descriptive statistics
  4. Optimization of different Machine Learning Models
  5. Reduction of information methods
  6. Agnostic methods
  7. Data augmentation
  8. Tuning of a pre-trained CNN model (Resnet-18)
  9. Comparative results

Key Features:

  • Data Integration: Combining both tabular data and image data for price prediction.
  • Model Comparison: Evaluating and comparing classical and deep learning models.
  • Model Interpretability: Using model-agnostic methods such as SHAP values and ICE plots to interpret the model results.

Dataset

The dataset used in this project is the DVM-Car dataset, which includes:

  • Numerical and Categorical Data: Attributes like mileage, engine size, fuel type, and gearbox type.
  • Image Data: Images of cars, including front-view images, which are used for CNN-based price predictions.

Data source : https://github.com/zalandoresearch/fashion-mnist

Data Preprocessing:

  • Data Cleaning: Handling missing values, outliers, and standardizing the data.
  • One-Hot Encoding: For categorical variables such as body type and fuel type.
  • Standardization: Applied to numerical features for models like KNN.
  • Data Augmentation: For image data, using random rotations and rescaling.

Model Architecture

K-Nearest Neighbors (KNN):

  • Applied to the tabular data with hyperparameter tuning to select the optimal number of neighbors.

Extreme Gradient Boosting (XGBoost):

  • A powerful tree-based model used for tabular data with a step-by-step hyperparameter tuning process.

Multilayer Perceptron (MLP):

  • A feedforward neural network applied to the tabular data, with extensive hyperparameter tuning to find the optimal network structure.

Convolutional Neural Network (CNN):

  • ResNet-18, a pre-trained deep learning model, was fine-tuned for car price prediction based on images.

Main Technologies

  • Python
  • Pytorch
  • Shap
  • Sklearn
  • XGBoost
  • Pandas
  • Google colab

This project is an assignment made in the context of my Master studies Aix-Marseille University Degree : Master in Econometrics, Big-Data & Statistics