Machine Learning-Based Detection of Driver Distraction by Capacitive Electrocardiogram Signals

Unobtrusive acquisition systems for recording vital parameters of drivers enable the development of ML models for monitoring health state, stress level, and driving conditions during driving in daily application. The paper proposed an ML-based model for the classification of four different driving conditions (driving without distraction, uncomfortable driving, distraction drivers by talking, and no driving) by the extraction of only five features from the capacitive electrocardiogram time series (cECG). The proposed method was validated on open data sets cECG recorded during simulated driving. The highest accuracy (95.83%), and Matthew's coefficient (0.95) were achieved by random forest classifier.