Time-series prediction of accident risk based on autonomic nervous function measured while driving

Abstract

Traffic accidents from the health condition of truck drivers is a social issue. We are developing a technology to detect danger in advance using physiological information. Our purpose is to analyze the changes over time in autonomic nervous function (ANF) estimated from heart rate variability analysis (HRV), and to predict the future risk of accidents while driving. During driving, it is necessary to adapt to a dynamic ANF environment different from the conventional static one. Therefore, we proposed a time series prediction method using two preprocessing and Deep Learning model. First, we extracted ANF index by continuous HRV with the optimized window width, and normalized data to eliminate individual differences. Then, environmental factors affecting ANF and accident risk were added to input. As results, Recall was 88.8% and AUC was 0.84. Even in during driving, ANF is effective for prediction of accident risk and it can contribute to accident prevention.