Feature Extraction Based on Wavelet Transform for Classification of Stress Level

We propose a methodology that classifies stress levels using a set of electrocardiography (ECG) signals collected from a web repository. Signals are first decomposed into ten multiresolution levels using a discrete wavelet transform (DWT). After noise removal, the features for detail coefficients are then generated by computing different statistical measures like mean, standard deviation, skewness, kurtosis, variance, root mean square, spectrum energy, Shannon entropy, log energy, form factor, and minimum and maximum value of wavelet coefficients. Dimensionality reduction via principal component analysis (PCA) occurs afterward. We finally invoked the support vector machine (SVM), weighted k-nearest neighbors (WKNN), and linear discriminant analysis (LDA) classifiers for stress level classification. Experimental results demonstrate that our proposed methodology as a whole is promising since the overall performance obtained with these three classifiers is around 98%. This work has proved to be a valuable tool to avoid the harmful consequences of human stress, (especially driver stress), in which an earlier and higher detection is necessary.