IoT-Based Multi-Sensor Fusion and Machine Learning Framework for Geo-Tagged Road Anomaly Classification

Road condition monitoring in real-time is important for safety, maintenance, performance, and smart transport systems. The conventional approach to road inspection involves manual inspections, inspection vehicles, cameras, or expensive sensors that prevent regular and large-scale monitoring. This research provides an IoT-based, multi-sensor fusion, and machine learning framework to detect and geographically locate road anomalies. The system employs an MPU6050 accelerometer and gyroscope, Global Positioning System (GPS), and Raspberry Pi 4 as a processor. Vehicle response data are collected in a continuous manner and further filtered, segmented, normalized, analyzed in terms of orientation, and features extracted. A Random Forest classifier is employed to classify roads into categories such as normal road, pothole, bump, and rough roads. Anomalies detected by the model are mapped to GPS coordinates for creation of geo-referenced road conditions data. Unlike vision-based monitoring, the proposed framework focuses on the analysis of vehicle dynamics and, therefore, is less dependent on lighting conditions and image processing requirements. Local processing also facilitates sustainable edge AI development by minimizing constant sending of data collected by sensors. The framework provides a practical approach to road health monitoring and further connected intelligent transport solutions