Comparison Of Machine Learning Algorithms For Heart Rate Variability Based Driver Drowsiness Detection

Abstract

Drowsy driving due to insufficient sleep has led to many serious traffic accidents. Measuring the drowsiness of the driver and taking timely actions can avoid such accidents. Earlier, conventional methods such as eye states and facial expressions were used to detect drowsiness. Nowadays new techniques have been developed for the same purpose, which uses bio-electric signals like an Electro Cardio Gram(ECG). Heart Rate Variability (HRV) can be used to assess drivers’ drowsiness, fatigue, and stress levels. HRV is determined by the interval of RR measured by an Electro Cardiogram. Twelve features are monitored, including both time and frequency domains, in order to determine the HRV changes. HRV monitoring is used to actually predict epileptic seizures. The proposed work uses Heart Rate Variability (HRV) analysis with a Machine Learning and Deep Learning to detect drowsiness. A comparison is also made between the performance of four different Machine Learning(ML) algorithms while using one-dimensional convolutional neural networks (1D CNNs). Convolutional neural networks (CNN) are used increasingly in Computer Vision and Machine Learning operations. 2D CNNs consist of millions of parameters and many hidden layers, and it has Interpreting complex patterns and objects. Two-dimensional signals, such as images and video frames, are used as inputs for 2D CNNs. However, this may not be the ideal choice in many applications, especially those involving One-Dimensional signals such as biomedical signals. To solve the problem, 1D CNNs were introduced with the highest level of performance. Specifically, the 1D CNN has four layers: a Convolutional Layer, Batch Normalization Layer, Maxpooling Layer, and Fully Connected Layer. The proposed strategy has the potential to help avoid accidents caused by drowsy driving.