Machine Learning Model for Anxiety Disorder Diagnosis Based on Sensory Time-Series Data

Anxiety disorders are common and serious mental disorders that affect many people and disrupt their lives. This study explores the potential of machine learning techniques in aiding the diagnosis of anxiety disorders through the analysis of sensory data. By leveraging features extracted from time-series bio-behavioral data, including motor activity and heart rate, predictive models are trained to differentiate between individuals with and without anxiety. Feature selection methods are employed to identify discriminative features for classification to enhance the predictive accuracy of the models. Three classification algorithms, Multilayer Perceptron (MLP), Support Vector Machine (SVM), and XGBoost are evaluated on a dataset of 68 samples from individuals with and without anxiety. The results show that the MLP classifier outperforms SVM and XGBoost with AUCROC = 0.94, precision = 0.87, recall = 1.0, and accuracy = 0.86 in classifying anxiety disorders from non-anxiety samples. This proposed machine learning model may serve as a potential diagnostics model for anxiety and may help understand the disorder’s relevant bio-behavioral and motor features.