Performance Analysis of Machine Learning Classifiers in Estimating the Driver’s Fatigue using Physiological Signals

Abstract: Driving fatigue is one of the significant factors that cause road accidents and often result in a huge socio- economic loss to the country. The accurate and reliable driver fatigue state assessment system can reduce the accident rate. In this proposed work, Heart Rate Variability (HRV) derived from Electrocardiography (ECG) is used as input to measure driver fatigue state. Machine learning classifiers like Support Vector Machine (SVM) classifier, Decision Tree, K-Nearest Neighbour (KNN) algorithm, Ensemble bagged tree classifier, Quadratic discriminant method and Deep Auto encoder techniques are used to estimate the driver fatigue state and their performance is also analysed. These machine learning classification systems use HRV features measured in time domain, frequency domain and also nonlinear HRV features. This study was conducted on 10 healthy individuals in simulator driving environment. The results have shown that deep auto encoder technique achieves highest accuracy of 97% in determining the fatigue level of drivers.

Keywords: ECG, HRV, Fatigue, SVM, KNN, Deep Auto Encoder

Objectives of the proposed work are to show that:  Heart Rate Variability (HRV) is one of the strong parameter to indicate the Fatigue Level  To acquire the ECG data from the subject using wearable ECG electrode and to calculate the heart rate using RR-Interval  To make use of the machine learning classifiers for estimating the fatigue level and also make the performance analysis of the classifiers used in the test.