Computational model for identifying stereotyped behaviors and determining the activation level of pseudo-autistic

Affective state recognition of an individual is based on the emotional cues, such as the activation level. Body expression is a modal able to convey emotions and can be used for autism diagnosis through the presence of stereotyped behaviors (SBs). These behaviors are atypical and repetitive movements of the body, which can be related to a low mental health condition. The development of systems able to both recognize SBs and inferring activation level can automatically aid some therapeutic approaches. In this paper, a computational model of low intrusiveness is proposed to infer activation levels from recognized SBs, Machine Learning Algorithms (MLAs) are for identifying the SBs and for determining the related activation levels. A metric performance is also proposed to evaluate the performance of MLAs considering the time for classification of the SBs, accuracy, and precision. For classifying the SBs, the Hidden Markov Models and Multilayer Perceptron presented the best performance than Support Vector Machine and Convolutional Neural Network. The Adaptive Neuro-Fuzzy technique based on the Fuzzy C-Means algorithm allowed one to determine and differentiate the activation levels of the stereotyped behaviors considered in the present study. The experiments were performed with non-autistic participants, here referred to as pseudo-autistic.