Comparative Prediction of Psychotic and Mood Disorders with Multi-Model Machine Learning

Recently, there has been a surge in the number of mental health cases including paranoid schizophrenia (psychosis) and depression (mood disorder). This study conducted a comparative prediction of psychotic and mood disorders using multi-model machine learning (MLs), mainly: Logistic Regression (LR), Support Vector Classification (SVC), Random Forest (RF), and Extreme Gradient Boost (XGBoost). Methods: The study used a clinical datasets of 318 patients diagnosed with Psychotic and Mood related disorders. We created a mid-category of Psychotic-Mood disorders for properly classifying patients having shared symptoms with Psychotic and Mood disorders. The Column Transformer preprocessor was used for modeling. The categorical columns were encoded using One-Hot-Encoder and Imputed using Simple Imputer with most-frequent as the strategy. The numerical columns were also encoded using Standard Scaler and imputed using Simple Imputer with mean as the strategy. Our results showed a consistent performance hierarchy in the confusion matrices (RF > XGBoost ≈ SVC > LR) for pure conditions, contrasted with the reversed pattern for mixed conditions, computationally validating long-standing clinical debates about psychiatric nosology and supports dimensional rather than categorical approaches, particularly highlighting that algorithmic complexity does not necessarily improve classification of inherently unstable diagnostic categories. We found that while conventional diagnostic criteria seeks to clearly differentiate psychiatric conditions, our computational evidence supports that psychotic-mood disorders often fall into a spectrum of conditions and that psychiatric comorbidity patterns can be detected using machine learning models.