MRMR Based Feature Selection For Multi-Class Mental Stress Detection With HRV Measures

Mental stress is a serious issue in today’s life that leads to physiological processes in the human body followed by many health problems and diseases. This paper presents a method for detecting multiple categories of mental stress based on the heart rate variability (HRV) measures derived from ECG signals. The minimum redundancy maximum relevance (MRMR) algorithm selects the most informative HRV measures to detect multi-class mental stress with less computation time. This paper applies different variants of decision tree (DT) and K-nearest neighbors (KNN) for training followed by comparative performance evaluation based on accuracy, training time, and error rate. The experiment is conducted on two well-known SWELL-KW and WESAD datasets considering 32 HRV unseen filtered measures in the time and frequency domain. Simulation results indicate that fine-KNN outperforms with an accuracy of 100% and zero error rate (validation cost) for both SWELL-KW and WESAD considering all features. Whereas, with selective five highly informative features achieved an accuracy of 99.95% for both; and 0.04% error rate (in SWELL-KW) and 4 validation costs (in WESAD). Fine-KNN with feature selection reduces the training time by approximately 40% and 31% in SWELL-KW and WESAD respectively.