An Explainable Machine Learning Approach for Early Heart Disease Prediction Using Noisy and Incomplete Healthcare Data

Cardiovascular disorders are among the primary causes of mortality worldwide, making
early detection a critical requirement in modern healthcare systems. Although machine
learning techniques have shown promising capabilities in disease prediction, their
performance is often affected by missing values, noisy records, and data inconsistencies
commonly found in clinical datasets. This study presents an explainable machine
learning-based approach for predicting heart disease using imperfect healthcare data. The
proposed method incorporates data preprocessing, feature optimization, ensemble
classification, and explainability mechanisms to improve both predictive performance and
transparency. Techniques such as SHAP and LIME are employed to interpret model
outputs and highlight feature contributions. Experimental results indicate that the proposed
approach achieves an accuracy of approximately 96–97%, outperforming baseline models
while offering meaningful explanations that can assist healthcare professionals in
decision-making.