Traffic Congestion Prediction

This repository contains the implementation of traffic congestion prediction, focusing on assessing and comparing the effectiveness of different algorithms for long-term traffic congestion prediction.

Objective

The primary objective of this project is to evaluate various algorithms in predicting long-term traffic congestion. Two datasets are used: one obtained from Kaggle and another generated from road traffic photos in Tirana. The data undergoes cleaning and preprocessing, incorporating a Convolutional Neural Network (CNN - YOLO V8) model for image processing and car counting. The YOLOv8 model was trained on a labeled car images dataset available at this link using CUDA and PyTorch.

Project Structure

• yolov8-car-detection: This file contains the model responsible for detecting cars in pictures, contributing to the creation of a time series dataset with images from Tirana.

• predictions/ARIMA: Folder containing predictions using the AutoRegressive Integrated Moving Average (ARIMA) model.

• predictions/GRU: Folder containing predictions using the Gated Recurrent Unit (GRU) model.

• predictions/LR_SVR: Folder containing predictions using Linear Regression (LR) and Support Vector Regression (SVR) with various kernels.

  • SVR_Kernels:
    • Linear Kernel:
    • Polynomial Kernel:
    • Sigmoid Kernel:
    • RBF Kernel:
    • Precomputed Kernel:
    • Custom Kernel:

Algorithms Used

Time Series Analysis:

• Autoregressive Integrated Moving Average (ARIMA)
• Gated Recurrent Unit (GRU)

Regression:

• Linear Regression (LR)

• Support Vector Regression (SVR) with various kernels.

  • Linear Kernel
  • Polynomial Kernel
  • Sigmoid Kernel
  • Radial Basis Function (RBF) Kernel
  • Precomputed Kernel
  • Custom Kernel

Note

The majority of labels in the output figures are in Albanian, as this was part of my master's dissertation project.