Optimal Time Segments for Stress Detection

Abstract

Some response signals being modeled for humans over some time segments may not be relevant for analysis and modeling. These signals could contribute to reducing the quality of patterns captured by models, inefficient processing and may impose huge demands on storage resources. This work proposes an approach to search for relevant time segments from human response signals particularly, physiological and physical signals to recognize stress. The paper proposes an approach to determine time segments that were critical to differentiate the types of text based on stress. A support vector machine (SVM) was used to classify the different types of text based on the features of the response signals. A SVM and genetic algorithm (GA) hybrid approach is developed to determine optimal time segments for stress detection (OTSSD). As well as optimizing time segments, the GA also dealt with hundreds of stress features that may have included redundant and irrelevant features. Optimal time segments for stress in reading were successfully found and the GA and SVM hybrid classifier showed an improvement in stress recognition when optimized features from the critical time segments were used.