An AI-Driven Statistical Learning Model for Medical Diagnosis: Integrating Bayesian Probability, Multivariate Analysis, and Predictive Machine Learning

Abstract

Medical diagnosis mediated by artificial intelligence demands dependable predictive models with the capacity for handling intricate diagnostic indicators and uncertainty in the form of probabilities. This study presents an integrated Bayesian–Multivariate–Machine Learning (BMML) framework which integrates the Bayesian inference, principal component analysis and classification of predictive models. It uses 1,000 observations, 10 diagnostic indicators, 2 diagnostic classes and 100 repeated evaluation runs in the computational design. Feature redundancy is dealt with by multivariate analysis, and logistic regression and random forest as well as gradient boosting aid in the diagnostic classification. Weighted probability fusion: A fusion of bayes and machine learning prediction. Here, the assumption of a likelihood ratio of 4.0 results in an illustrative Bayesian calculation that increases the posterior diagnostic probability from 30% to 63.16%. The evaluation framework also takes into account the metric of accuracy, sensitivity, specificity, F1-score, ROC-AUC and calibration reliability. The proposed approach will lay a structured framework for reviewing diagnostic discrimination and uncertainty, and it is important to remember that its effectiveness needs to be proven clinically in the future.