A Novel Approach for Classification of Proneness of Schizophrenia in Patients using Ensemble Machine Learning Techniques and LIME

In the present scenario, where physical disorders come to light in no time, mental disorders such as Schizophrenia (SZ) often get neglected due to inefficient current clinical methods. The reason being the huge amount of time taken between the very first symptom and the definite diagnosis of this disease. Schizophrenia affects approximately 24 million people or 1 in 300 people (0.32%) worldwide [1]. When compared to the existing methodologies for the prediction of Schizophrenia which involve medical professionals observing demeanor, and asking about thoughts, moods, delusions, hallucinations, etc., the Machine Learning (ML) approach used in this work along with LIME results, is potentially less time-consuming for early detection of the susceptibility to Schizophrenia. Also, as compared to methodologies like MRI or CT scan, which are used to rule out conditions with symptoms similar to Schizophrenia, this ML approach can potentially be cost-effective. This study uses Machine Learning classification algorithms and ensemble techniques to classify the proneness of Schizophrenia in patients. The ensemble technique, stacking, is used in this work. This ensemble technique considers heterogeneous weak learners and combines them to produce predictions. The k-fold cross-validation technique yields a less biased model, ideal for limited data. The ensemble model used in this work, comprising of XGBoost (Extreme Gradient Boosting) Classifier, Decision Tree Classifier, Random Forest Classifier and Linear Discriminant Analysis, and Logistic Regression as a meta-classifier, has shown a cross-validation score of 0.9602 (96.02%). The Explainable AI (XAI) technique and LIME (Local Interpretable Model-Agnostic Explanations) are used for model interpretability. To improve the quality of life and outcomes for those who are affected by Schizophrenia, this work aims to increase the accuracy and efficiency of prediction of Schizophrenia. The proposed diagnostic and predictive rules provide a foundation for the development of decision-support systems in clinical settings.