NeuroInk: Transparent AI for Mental Health Assessment Using Handwriting Dynamics

The detection of schizophrenia (SZ) and bipolar disorder traditionally relies on expensive imaging techniques such as MRI, which can limit accessibility and patient compliance. To overcome these challenges, this study proposes a cost-effective handwriting-based analysis approach that leverages motor abnormalities commonly associated with these disorders. The DataRepository SAV dataset from Figshare is used, containing handwriting-derived features for classification. Data preprocessing includes robust feature selection using Recursive Feature Elimination (RFE) and handling class imbalance through cost-sensitive learning with class weighting. Several machine learning models are implemented, including XGBoost, Logistic Regression, Linear Discriminant Analysis (LDA), Naïve Bayes, K- Nearest Neighbors (KNN), and Support Vector Machines (SVM with linear and RBF kernels). An ensemble Voting Classifier combining XGBoost and Logistic Regression is also developed to improve predictive performance. Model evaluation is conducted using accuracy, precision, recall, F1-score, Matthews Correlation Coefficient (MCC), and confusion matrix. The ensemble model achieves the highest test accuracy. For interpretability, SHAP-based Explainable AI (XAI) techniques are applied to identify key contributing features. Finally, the optimized model is deployed through a Flask-based web application for real-time and transparent patient classification.