Abstract:
The escalation of road accidents and associated fatalities presents a pressing challenge, particularly in developing nations like Bangladesh. This study addresses the cri...Show MoreMetadata
Abstract:
The escalation of road accidents and associated fatalities presents a pressing challenge, particularly in developing nations like Bangladesh. This study addresses the critical need for enhanced road safety by introducing a sophisticated “RoadSense” device that integrates sensor technology and machine learning to predict road conditions in real time. The system employs an MPU-6050 IMU module along with GPS, temperature and humidity sensors to collect data in real-time during vehicle operation. The data collection of the device is complemented by training of tree-based ensemble supervised models which are refined with the SMOTE-ENN technique for class imbalance of data. Post-hyperparameter tuning, the Extra Trees Classifier achieved an accuracy of 99.07%, with a recall of 97.17% and precision of 98.66% for pothole detection, and similarly high metrics for detecting speed breakers and hard braking events. In real-world testing, the model maintained a high F1 score of 97.57% across diverse road conditions. Unlike existing solutions, the RoadSense device uniquely combines low-cost components, multi-sensor hardware development tailored for developing countries with advanced algorithm-level data processing techniques. This integration delivers a scalable, real-time road monitoring solution that supports data-driven infrastructure maintenance and policy decisions in. A key distinguishing feature of this work is that the entire framework is specifically designed to be both implementable and scalable within the context of developing countries such as Bangladesh.
Published in: IEEE Transactions on Intelligent Vehicles ( Early Access )