Evaluating Machine Learning and Association Rule Techniques for Health Data Mining: A Comparative Study on KNN, Naïve Bayes, and Apriori Algorithms

Abstract

Accurate disease classification and behavioral pattern mining are crucial for early intervention and preventive healthcare. While machine learning models have been extensively applied in health informatics, research that compares the performance of K-Nearest Neighbors (KNN), Naïve Bayes (NB), and Apriori algorithms across various health datasets remains limited. This study evaluated the performance of KNN and NB classifiers on five Kaggle datasets covering lung cancer, heart disease, depression, diabetes, and cardiac risk. The Apriori algorithm was used for mining association rules in the Depression dataset. Preprocessing included data, scaling, and dimensionality reduction using Principal Component Analysis (PCA) to improve efficiency. KNN demonstrated more reliable performance than NB across datasets, achieving an average accuracy of more than 0.92, especially in numeric-heavy environments. The study also identified lifestyle factors significantly associated with depression risk through Apriori mining. The findings highlight the strong potential of lightweight machine learning models for real-time health monitoring, early diagnosis, and behavioral intervention applications.