Research on anxiety detection based on personalized data markers

Abstract

The use of machine learning methods to detect human anxiety has become the mainstream. However, the label of current data usually relies on the subject's manual labeling, and there are strong subjective factors. In addition, traditional methods have the disadvantages of insufficient feature extraction, inaccurate classification results, and insufficient generalization ability. In response to the above problems, we propose an objective data labeling method, which extracts features on the basis of maintaining the characteristics of the original physiological signals, constructs a more complete feature set, and builds a long short-term memory network (LSTM) model on this basis perform an anxiety state detect. Experiments show that after processing the data by our method, the performance of all models has been greatly improved, among which LSTM shows the best performance compared with traditional algorithms.