Utilizing Biometrics to Recognize Emotions in Individuals with Autism Spectrum Disorder

Over 75 million people worldwide have Autism Spectrum Disorder (ASD). Up to 30% of children with ASD are minimally verbal, making it hard to recognize their emotions and leading to preventable meltdowns. To address this, a cost-efficient wearable device is designed to predict emotions by capturing and classifying biometric heart rate signals. Using a Max30100 sensor, the device collects photoplethysmography (PPG) data, which is pre-processed and analyzed to extract heart rate variability (HRV) features for emotion detection. A bright-colored glove secures the electronic components and engages children with ASD. The device uses a K-nearest neighbors classifiers machine learning (ML) model with 7 output classes (7 different emotions) trained and cross-validated on the YAAD dataset. The algorithm had up to $\mathbf{6 8. 4 2 \%}$ accuracy in classifying between the $\mathbf{7}$ emotions. Future improvements include expanding the dataset, training on self-recorded signals, and integrating a real-time wireless app.