A study on reliability-oriented composite traffic networks for urban agglomerations

Urban traffic conditions are affected by temporal variations, environmental factors, traffic volume, and uncertain operating conditions, making reliable accident occurrence identification challenging. This study proposes a Sensitrix Global Sensitivity-Based Traffic Accident Classification framework that integrates Min–Max Scaling, Discrete Wavelet Transform (DWT), Monte Carlo Simulation (MCS), Global Sensitivity Analysis (GSA), Fossa Optimization Algorithm (FOA), and Long Short-Term Memory (LSTM) classification. The preprocessing and analysis stages transform traffic-related observations into an informative feature representation, while the LSTM model performs binary classification of traffic conditions into Accidents and No Accidents. The proposed framework is evaluated using accuracy, precision, recall, F1-score, NPV, MCC, ROC-AUC, and PR-AUC. The experimental results demonstrate 99.88% accuracy, 98.88% F1-score, 0.9984 PR-AUC, and 0.9988 ROC-AUC, indicating effective discrimination between accident and non-accident traffic conditions.