Deep neural network-based stress detection using biosignals

Recently, electrocardiogram (ECG) signals have been used in affective computing to recognize and interpret human emotions, loneliness, pain, and other psychological states. In healthcare, human-computer interaction, and good health, stress detection is one application of affective computing that has been gaining attention. As one of the major factors in diagnosing stress, artificial intelligence uses sophisticated algorithms to examine physiological data that is difficult to interpret and find subtle signs of stress. Our article proposes a stress prediction model based on long short-term memory (LSTM). In this model, the temporal characteristics of the data are processed using recurrent neural networks, which are compatible with time-series data such as ECG signals. First, the ECG signals are normalized, and then the input is classified further to assess the performance of the proposed model. The features are extracted using a four-layered LSTM network and wavelet transform. Afterwards, the features are optimized to improve classification accuracy. Due to its high sensitivity and accuracy, the proposed model outperforms other methods of feature extraction such as kernel methods, principal component analysis, and embedders.