abstract
- This study addresses the need for advanced road assessment methodologies in Intelligent Transportation Systems (ITS). Our investigation explores Deep Learning (DL) methodologies to enhance road condition prediction accuracy by analyzing vehicle vibration data. Through strategic positioning of Inertial Measurement Units (IMUs) within vehicles, our research aims to categorize road conditions into five distinct classes containing various anomalies and elements of the road. By proposing and testing 1-dimensional Convolutional Neural Network (CNN) architectures and diverse data inputs, we refine the classification process to optimize performance. We conduct a comparative analysis between DL models and traditional algorithms, highlighting the advantages of DL methodolo-gies. Notably, our DL models surpass traditional algorithms by increasing overall classification performance by 0.103, as evidenced by F1-scores obtained in our experiments. This study stands as a significant contribution to the field of ITS, offering and facilitating more precise and reliable road condition assessments. Key contributions include the development and assessment of tailored DL models for road condition classification, as well as comparative analysis with traditional algorithms. Through this research, we underscore the potential of DL techniques in significantly improving road condition prediction accuracy, thereby driving the evolution and success of ITS applications. © 2024 IEEE.