Optimal Decision Tree for Early Detection of Bipolar Disorder based on Crowdsourced Symptoms

Bipolar disorder is a chronic mental health disorder identified by periodic manic or depressive episodes. Early intervention for bipolar disorder is necessary to prevent progression and complications that lead to more societal loss. In this study, we build an early detection model for bipolar disorder based on crowdsourced mental health symptoms. The mental health symptoms are gathered through the crowdsourcing process in the form of free texts. The feature extraction is done using natural language processing techniques to convert free texts into binary features. Based on these features, we build an optimal decision tree model by formulating a mathematical optimization problem that minimizes misclassification loss and penalizes the number of leaves in the tree, constrained by a depth bound. The optimal decision tree model outperforms the baseline models in terms of accuracy (0.899), recall (0.869), precision (0.921), F1 score (0.894), and AUC (0.898). Moreover, the model is interpretable since it maintains the tree-like structure as in other decision tree models. This model can be used as an early detection tool to recommend for further examination of diagnosing bipolar disorder.