Feature Selection based on Evolutionary Algorithms for Affective Computing and Stress Recognition

In the area of affective computing, machine learning is used to recognize patterns in datasets based on extracted features. Feature selection methods are used to select the relevant features from the large number of extracted features. This paper presents a feature selection approach based on evolutionary algorithms using techniques inspired by natural evolution, such as inheritance, mutation, selection and crossover. Our proposed method consists of the steps Initialize, Evaluate, Mutate, Crossover and Select. First, an initial population consisting of a set of individuals is generated, in which every individual has a randomised set of features. Then, the fitness of every individual representing the accuracy of the prediction is evaluated to select the fittest individuals for the next steps Mutation and Crossover. Mutation sets used attributes to unused atrributes and vice versa, while with Crossover one part of one individual is crossed over with another part of the other individual. Finally, the performance of the new individuals is calculated and they are pitted against each other. Individuals with the higher performance have a higher chance to survive for the next round. The feature selection method with evolutionary algorithms is integrated within our previously developed workflow for affective computing and stress recognition from biosignals and is evaluated using the University of Ulm Multimodal Affective Corpus (uulmMAC) for Affective Computing in Human-Computer Interaction. Our proposed approach is much faster than the Forward Selection (FS) and Backward Elimination (BE) methods and does not stop at a local optimum, allowing a promising feature selection alternative in the field of affective computing.