An Empirical Study of Machine Learning Models to Identify Human Stress Levels using Brain Signals

Abstract

In human life, stress starts at an early age, it can either be positive or negative. Con- newlinetinuous failure to achieve targets and a hectic work schedule are the main reasons for newlineincreasing stress. Negative psychological stress for a long time affects daily life and newlinecauses many diseases. Many different types of biological signals, including tempera- newlineture, electrical conductance, impedance, acoustic volume, and optical clarity, all fluctu- newlineate in response to stress. So, early detection of stress is important for saving human life. newlineElectroencephalogram (EEG) signal recording instruments are commonly employed for newlinethe purpose of collecting the brain patterns in the form of electric waveforms. In the newlineproposed framework the EEG signals have been used as inputs. During the recording newlinetime original signals mixed with unwanted signals called as noise. To extract needed newlinedata signals from dataset, data preprocessing techniques play an important part. newlineThe EEG signals dataset collected from well-known Physionet data repository to newlinedetect the stress levels. In this dataset, the brain EEG is collected when subjects per- newlineform arithmetic calculation tasks. Discrete Wavelet Transform (DWT) has been used newlineto filter out noise and divide multichannel EEG signals into five distinct frequencies newlineband. From the distinct frequency band signals, three feature selection methods are newlineused: combination of statistical importance (CIS), an automated subset of feature se- newlinelection method Convolution Neural Network (CNN), and a feature subset optimized by newlinecuckoo search optimization (CSO). After preprocessing and features selection different newlineclassification techniques are proposed to detect human mental stress levels.