Combining multiple stress identification algorithms using combinatorial fusion

摘要

Sensor feature selection and combination algorithms are important in the identification of stress level for human activities and health. However, performance of each algorithm may depend on physiological sensors, sensing modalities and feature selection methods. Based on our previous work on feature selection and combination, we study combination of five sensor stress identification algorithms (SIA): C4.5, Naive Bayes, Linear Discriminant Function, Support Vector Machine and K Nearest Neighbors across a variety of feature sets selected by C4.5, PCA, Correlation-based Feature Selection (CFS) and Diversity-based Feature Selection (DFS). Our experimental results demonstrate that combinatorial fusion is a viable method to improve identification of stress in human activities and health. Moreover, we observe that the improvement is stronger when the cognitive diversity between individual algorithms is bigger.