Machine Learning-Based Road Hazard Detection for Improved Driving Assistance and Safety

Road hazard detection systems are essential for identifying surface anomalies such as speed bumps, cracks, potholes, and damaged road segments, thereby helping drivers adjust their speed and improve overall safety.This research introduces a novel deep learning-based architecture, Particle Swarm Recurrent Sequential Memory NeuroNet (PSRSMN), designed to enhance the accuracy and robustness of road anomaly detection.The Kaggle Bump Detection dataset, encompassing real-world road conditions such as bumps, cracks, potholes, and damaged roads, is employed to train and evaluate the model.The methodology begins with a preprocessing stage involving image resizing and color space transformation, improving visual consistency and preparing the data for optimal feature learning.The core innovation lies in the integration of Local Binary Pattern (LBP) for texture-based encoding and EfficientNet for deep spatial feature extraction, which together reduce computational load while preserving critical structural information.To address temporal dependencies often ignored in standard approaches, a recurrent sequential memory module is embedded into the network, enabling the capture of contextual relationships across frames or segments.Crucially, Particle Swarm Optimization (PSO) is employed to automatically fine-tune hyperparameters across the network components, ensuring efficient convergence and avoiding manual tuning bottlenecks.This hybrid design, combining handcrafted texture features, deep learning, memory-based reasoning, and Particle Swarm Optimization-based swarm intelligence, forms the basis of the model's novelty and distinguishes it from conventional Convolutional Neural Network (CNN)-only methods.Using the Kaggle Bump Detection dataset, the proposed PSRSMN model achieves 99.37% accuracy, 99.49% precision, 99.41% recall, 99.39% F1-score and [email protected] at 99.98%, outperforming baseline models in both performance and stability.These results validate PSRSMN as a reliable, real-time diagnostic framework for road hazard detection, contributing to improved transportation safety and proactive traffic infrastructure monitoring.