Sense-of-agency as clinically accessible features for schizophrenia prediction: Interpretable ensemble machine...

Schizophrenia is a high-risk, high-burden psychiatric disorder characterized by a prolonged course and severe disability. Accurate identification and early intervention can mitigate socioeconomic adversities, including illness-induced poverty and public safety risks. Traditional diagnosis relies predominantly on structured clinical interviews, with clinicians using positive symptoms as key diagnostic indicators for schizophrenia and related disorders. Recent advances in machine learning-driven computer-aided diagnostic systems have emerged as a transformative frontier. These systems can effectively capture correlations between quantifiable features and schizophrenia, enabling not only auxiliary diagnostic predictions but also providing potential directions for clinical treatment. Notably, deficits in sense of agency (SoA) represent a core feature of schizophrenia; however, directly predicting schizophrenia based on SoA deficits and quantifying their diagnostic significance remain critical unresolved challenges. In this work, one hundred and fifty-five participants were recruited, and interpretable ensemble machine learning models were developed to investigate the SoA features for schizophrenia prediction and interpretability analysis. First, agency rating, time interval estimation and intentional binding methods were used for SoA features generation. Then, six baseline machine learning algorithms were trained, with RF and TabPFN demonstrating optimal performance. To further enhance reliability, an ensemble modeling strategy with RF and TabPFN was implemented, yielding a high-performance classifier SchNet (Accuracy of 0.90, F1-Score of 0.91). To bridge theory and practice, we also deployed SchNet webserver (https://github.com/jourmore/SchNet-webserver), offering SoA test online, schizophrenia risk prediction and interpretability analysis. This tool serves as a translational bridge between computer research and clinical application, supporting data-informed therapeutic strategies for schizophrenia.