Stress and Emotion Detection System Using IoT and Machine Learning with a Fuzzy Inference-Based Mental Health Risk Assessment Module

Abstract

This research paper describes the conceptualization, deployment, and application of a stress and emotion detection system (SEDS) that utilizes emerging technologies including machine learning, the Internet of Medical Things (IoMT), and fuzzy logic. Particle Photon microcontrollers were used to collect and analyze data from non-invasive biosensing devices, input switches for behavioral, emotional, and cognitive stress indicators, and input signals generated by the machine learning-based facial identification sub-module, which detects an individual's emotional states (happy, sad/upset, angry/irritable, or nervous/scared). The acquired parameters were stored and made readily accessible using IoT cloud services and dedicated mobile phone applications. The emotional condition of an individual was assessed by the utilization of the Personal Image Classifier tool within the MIT App Inventor, which involved the analysis of facial expressions. Furthermore, the SEDS was integrated with a module for mental health risk assessment which employs fuzzy logic to categorize the user's stress-related psychological health threat level as very low, low, moderate, high, or extremely high based on the data collected. The customized smartphone application provided users specific recommendations for effectively managing their mental health, based on stress level assessments. When the SEDS determined that the level of mental health risk posed by stress was high, it automatically generated a referral notification and transmitted it via text message to the mental health care professional, facilitating the provision of appropriate psychological counseling. Based on the collected data, the stress and emotion detection system produced results that were comparable to those from the DASS21 stress scale. The system demonstrated an improved accuracy of 90% in a test involving thirty individuals who volunteered to participate. The machine learning-based emotion detection system achieved a classification accuracy of 86.67% in correctly detecting happy, neutral, sad, angry, or nervous feelings through the analysis of facial expressions. This research is designed to provide mental health care professionals such as guidance counselors, psychologists, and psychiatrists with the resources essential to facilitate the evaluation and treatment of mental health issues. It also aims to raise people's understanding and detection of psychological conditions through enhanced awareness initiatives.