Analyzing road conditions in electric vehicle driving using machine learning and canbus data

Abstract

This study proposes a sensor-free approach for road condition detection using CANbus data from a battery electric vehicle. Real-world driving data were collected under similar driving behaviors across four road types: asphalt, concrete, gravel, and bumpy surfaces. Descriptive metrics such as Pearson correlation, Jerk, and Root Mean Square (RMS) were used to analyze the relationship between driver input and vehicle response. The correlation between pedal position and acceleration reached 93% on asphalt, 92.7% on concrete, 90% on gravel, and 87% on bumpy roads, indicating a strong influence of surface condition. Additionally, wheel slip variability was found to be a key indicator of road quality. A feature set was derived from standard CANbus parameters-speed, acceleration, and motor torque-without requiring any additional sensors. Machine learning models including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) were trained for classification. Among them, SVM achieved the highest accuracy (95.18%) in the shortest training time. The proposed approach demonstrates high performance and practical applicability for real-time road condition estimation using existing in-vehicle data.