DSGP-GROUP-9: UK Road Accident Analysis & Prediction

A Data Science Group Project (DSGP) that analyzes historical UK road accident data and serves machine-learning predictions through a Flask web application. The system helps identify accident hotspots, predicts accident severity, and visualizes accident trends for several UK cities.

Features

• Accident count / hotspot prediction — predicts expected accident counts for a given month, hour, day of week, district, weather, and lighting condition, and highlights the location on a map.
• Road surface condition prediction — predicts the likely road surface condition (dry, wet/damp, frost/ice, snow, flood) based on weather, light, and road type.
• Accident severity prediction — predicts severity (Fatal / Serious / Slight) based on number of vehicles, number of casualties, weather, road surface, and lighting conditions.
• Analysis dashboard — interactive charts (road type distribution, speed limit distribution, accidents by day of week) for cities such as Liverpool, Knowsley, Manchester, Leeds, Birmingham, Westminster, Bradford, Kirklees, Sheffield, and Leicester.
• Login / sign-up page as the entry point to the application.

Repository Structure

DSGP-GROUP-9/
├── connecting model/          # Flask web application
│   ├── main.py                # Flask app: routes, model loading, mappings
│   ├── *.pkl / *.pkl.gz       # Trained ML models (pickled)
│   ├── templates/             # Jinja2 HTML templates
│   │   ├── loginpage.html
│   │   ├── homepage.html
│   │   ├── aaa.html           # Accident count / hotspot prediction page
│   │   ├── bbb.html           # Severity prediction page
│   │   └── analystpage.html   # Analysis dashboard
│   └── static/                # CSS, JS, images, map templates
├── *.ipynb                    # Jupyter notebooks for data cleaning, EDA,
│                               # and model training
├── Birmingham_road_surface_dataset.csv
├── newDataset.csv
└── Local_Authority_Districts_December_2022_UK_BGC_V2_.../  # UK district
                                                              # boundary shapefiles

Tech Stack

• Backend: Python, Flask
• Machine Learning: scikit-learn (Random Forest models), pandas, numpy
• Data Analysis: Jupyter Notebook, pandas, numpy, matplotlib, seaborn
• Frontend: HTML, CSS, JavaScript, Google Charts, Leaflet/Folium (maps)
• Data: UK road accident datasets (Accidents/Vehicles 2005–2014, Birmingham road surface dataset, custom processed datasets), tracked with Git LFS

Getting Started

Prerequisites

• Python 3.10+
• pip

Installation

pip install flask numpy pandas scikit-learn

Running the Web App

cd "connecting model"
python main.py

The app starts on http://localhost:8000/. Navigate to the login page, then use the PREDICTION, SEVERITY, and ANALYSIS links in the navigation bar to access each feature.

Notebooks

The root-level .ipynb files contain the data cleaning, exploratory data analysis, and model training steps used to produce the pickled models in connecting model/. Open them with Jupyter Notebook or JupyterLab to review the analysis.

Models

File	Purpose
best_random_forest_model_zip.pkl.gz	Accident count / hotspot prediction
road_surface_model.pkl	Road surface condition prediction
random_forest_model.pkl	Accident severity prediction
accident_severity_model.pkl	Severity model (alternate)

Data

Large datasets (Accidents0514.csv, Vehicles0514.csv, etc.) are tracked using Git LFS — make sure git lfs install has been run before cloning to retrieve these files.