This document discusses using different machine learning models like regression, neural networks, decision trees, and memory-based reasoning (MBR) models to predict heart disease using a dataset containing medical information from 89,000 patients. MBR models use k-nearest neighbors to predict values based on similarity to past cases. The author finds that MBR models performed best on this task with low error rates and high lift, suggesting MBR is well-suited for clinical prediction using past patient data. Key factors for heart disease identified are age, blood pressure, cholesterol, smoking, and body fat percentage.