excelasaph/Predicting-Road-Crash-Severity-in-Nigeria-Using-ML-Optimization-Techniques

# Predicting Road Crash Severity in Nigeria Using ML Optimization Techniques - This project tackles Nigeria's high road traffic fatality rate (41,709 annually) using FRSC Road Transport Data from the National Bureau of Statistics - The dataset includes 481 records with causative factors (e.g., speeding 56.3%, tire bursts) and crash severity (fatal/non-fatal) across Q3 2021–Q3 2024. - It implements optimized ML models (SVM, XGBoost) and neural networks with regularization to predict severity, addressing class imbalance with SMOTE. - The goal is to improve FRSC interventions, aligning with Nigeria’s National Road Safety Strategy II. **Project Scope:** I first started out with cleaning and engineering this data into 8 percentage-based features (e.g., `SPV_PCT` for speeding), addressing a severe 477:4 class imbalance with SMOTE, and developing a suite of machine learning models like Support Vector Machines (SVM), XGBoost, and neural networks with regularization. ## Project Structure ``` ├── data/ │ └── Road Transport Data Q3 2024.xlsx ├── model_architecture/ │ └── Road Traffic Model Architecture.png ├── saved_models/ │ ├── no_optimization_model.keras │ ├── optimized_nn1_model.keras │ ├── optimized_nn2_model.keras │ └── xgboost_best_model.pkl ├── Summative_Intro_to_ml_[Excel_Asaph]_assignment.ipynb └── README.md ``` ## Dataset The dataset is derived from the **FRSC Road Transport Data** collection, accessible via the National Bureau of Statistics Microdata Catalog (microdata.nigerianstat.gov.…). It is available for download as an Excel file here: `data/Road Transport Data Q3 2024.xlsx`, with the latest update on May 09, 2025 - **Features and Target:** - **Crash Data Features:** - `FATAL`: Number of fatal crashes. - `SERIOUS`: Number of serious crashes. - `MINOR`: Number of minor crashes. - `TOTAL CASES`: Total crash incidents. - `NUMBER INJURED`: Number of injured individuals. - ` …