Affective state classification using timing-related features from short windowed PPG signal

The aim of the study is to verify the possibility of accurately recognizing an individual's affective state through the measurement of parameters extracted from a single PPG signal. For this, we used the WESAD public database consisting of a set of multidomain physiological signal. Specifically, the PPG was chosen as non-invasive signal, easy to collect and indicative of the cardiovascular system's response to the subject's emotional involvement. The considered affective state consisted of three conditions: baseline, stress state and amusement state. 17 timing-related features were extracted from the windowed PPG signal and a SVM and a k-NN classifiers were used to predict the affective state of the subject. This study proved the efficacy of the features extraction algorithms related to the time domain, reaching an accuracy of approximately 57% on the test data for both the classifiers. Our work highlighted that the PPG signal analysis combined with appropriate features extraction methods and classification models can lead to good results in detecting the affective state of a subject without considering the frequency domain features.