Machine Learning Model for Road Anomaly Detection Using Smartphone Accelerometer Data

This paper presents a vibration-based machine learning approach for road surface monitoring using smartphone sensors. With Mexico’s road network experiencing significant deterioration and potholes ranking as citizens’ top concern, we propose a convolutional neural network (CNN) model that analyzes accelerometer and gyroscope data from Android smartphones to detect road anomalies. Our methodology includes a custom mobile application for data collection, feature extraction through moving average filtering, and a 2-CNN architecture for classification. Experimental results demonstrate 98% accuracy in distinguishing potholes from speed bumps when using six sensor features, compares favorably with previously reported vibration-based approaches. The system’s low-cost implementation and high accuracy indicate that it may be well suited for large-scale road condition monitoring using mobile crowd-sensing paradigms.