Platelet Count as a Predictive Biomarker for Schizophrenia

Abstract

Schizophrenia is a severe psychiatric disorder whose diagnosis remains largely dependent on clinical assessment, underscoring the need for objective and interpretable decision-support tools. This study investigates the use of platelet-based biomarkers and demographic variables for predicting schizophrenia through machine learning and explainable artificial intelligence. A structured dataset containing 151 observations with gender, age, platelet count, and diagnostic class was analyzed using a stratified train-test split. Nine classification algorithms were evaluated, including Logistic Regression, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, K-Nearest Neighbors, Support Vector Machine, XGBoost, and LightGBM. Model performance was assessed using accuracy, precision, recall, F1-score, and receiver operating characteristic metrics. XGBoost achieved the best overall performance, with an accuracy of 0.652, recall of 0.773, F1-score of 0.680, and ROC value of 0.643. To improve clinical interpretability, SHAP and LIME were applied to examine global and local feature contributions. Both methods consistently identified platelet count and age as the most influential predictors, while gender had a comparatively smaller contribution. These findings suggest that explainable machine learning may provide a transparent framework for supporting schizophrenia prediction and for identifying potentially relevant biological markers. Further validation using larger, clinically diverse datasets is required before practical clinical deployment.