Outlier Detection For Robust Medical Diagnosis: A Benchmark Evaluation Across Classifiers

Abstract

Outliers in datasets can substantially affect the efficacy of machine learning classifiers, particularly in sensitive domains specificaly healthcare. This study examines multiple outlier detection techniques on benchmark medical datasets, including heart disease, diabetes, and breast cancer. Seven methods spanning statistical, clustering, and proposed approach were applied to each dataset, after which classification was performed independently using Random Forest, Logistic Regression and Support Vector Machine. The results show substantial improvements to the model’s resilience and accuracy. Notably, all classifiers achieved 100% accuracy on the heart disease dataset. Random Forest attained 99.07% accuracy on both Wisconsin Breast Cancer Dataset and the Pima Indian Diabetes Dataset.