Advanced Weather-Driven Predictive Modeling of Road Traffic Accidents in Nigeria Using State-of-the-Art AI Techniques

Abstract:

This research presents a comprehensive approach to understanding and predicting road traffic accidents in Nigeria based on weather data. By leveraging advanced machine learning and deep learning techniques, the aim is to create a system that forecasts both accident occurrences and severity with high accuracy. Using methods such as Transformer-based time-series forecasting, ensemble learning, and Explainable AI (XAI), the research explores how environmental factors impact road safety. Key insights, such as variance explained through feature analysis and the role of specific weather variables, will contribute to actionable policy interventions.

Research Objectives:

1. Predictive Modeling for Road Traffic Accidents: Build a sophisticated model that predicts the occurrence of road traffic accidents based on weather conditions, utilizing high-dimensional features and temporal trends.
2. Accident Severity Classification: Develop models that assess the severity of crashes in terms of injuries, fatalities, and vehicles involved, correlating them with weather data.
3. Real-Time Risk Scoring System: Construct a dynamic risk-scoring model that adapts to real-time weather conditions to provide actionable insights for authorities and the public.

Key Metrics and Results Incorporated:

1. Explained Variance Ratio:

• PCA Explained Variance: The first three principal components explain 99.88% of the variance in the weather dataset, with the breakdown as follows:
• PC1: 65.77%
• PC2: 33.28%
• PC3: 0.84%

These results indicate that the weather data can be significantly compressed without losing much information, allowing for more efficient modeling.

2. Mutual Information Between Features and Crash Data:

• Humidity shows the highest mutual information with crash data, at 0.72, suggesting that it is a critical factor influencing road accidents.
• Temperature has a…