Explainable Artificial Intelligence for Psychiatric Illness Prediction: An Interpretable Machine Learning Approach

Abstract

The early detection of psychiatric disorders is still difficult, because of the complexity and subjectivity of clinical diagnosis, and psychiatric disorders are a significant public health problem worldwide. This study suggests an Explainable Artificial Intelligence (XAI) machine learning system to accurately predict psychiatric illnesses with explanation. In this study, data from a psychiatric healthcare data set was processed using data cleaning, feature selection, and data normalization techniques in a retrospective manner in India. Five machine learning algorithms were created and tested: Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM) and Extreme Gradient Boosting (XGBoost). Confusion matrix analysis, AUC-ROC, accuracy, precision, recall and F1-score were used to evaluate the performance of the model. Automatic detection of schizophrenia with XAI can help patients by facilitating early intervention, helping physicians make accurate diagnoses, and helping medical staff maximize resource allocation and care coordination. Machine learning (ML), a subset of AI, is crucial since novel methods are required due to its complexity. ML technologies are unique in their capacity to assess a wide range of data, including genetic, neuroimaging, and clinical evaluations. The AI/ML literature in mental health lacks a consensus on the meaning of explainability. In XAI, it generally refers to model-agnostic techniques that make complex models more understandable for humans through simpler, interpretable explanations. An AI is opaque or "black-box" when the computational mechanisms that intervene between an input and the AI's output are too complex to provide a basic explanation of why the model produced that output—the exemplar case being deep neural networks, where computational complexity provides remarkable flexibility at the expense of increasing opacity.