A Comparative Analysis of Dataset Performance in Disease Prediction via Machine Learning Algorithm

Abstract

This study aims to investigate how dataset characteristics influence the predictive performance of machine learning (ML) algorithms in the context of disease diagnosis. While existing literature often focuses on evaluating the performance of various models on a single dataset, this study adopts a broader perspective. The UCI Heart Disease, Heart Failure, and Cleveland datasets were pre-processed using various techniques to ensure structural comparability and subsequently analyzed using models developed with the CatBoost algorithm. The study assesses the performance of these models on each dataset and explores the influence of different parameters. The model demonstrated strong predictive capability across all datasets, achieving high accuracy scores. For the UCI Heart Disease dataset, the model was able to effectively distinguish between classes, supported by an accuracy rate of 84.24% and other performance metrics. On the Heart Failure dataset, the model exhibited even higher performance, with an accuracy of 88.59%. The Cleveland dataset also yielded favorable results, achieving an accuracy of 85.25%. The results underscore the practical value of ML-based classifiers in the early prediction of heart-related medical conditions. By comparing model success across different datasets, the study highlights the applicability and effectiveness of these techniques and provides direction for futureresearch involving larger datasets and alternative algorithms.