ON THE GENERALIZATION GAP IN MACHINE LEARNING MODELS FOR HEART DISEASE PREDICTION: A CROSS-DATASET STUDY
International Research Journal of Modernization in Engineering Technology & Science · 07.06.2026
DOI: 10.56726/irjmets101546 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Research Journal of Modernization in Engineering Technology & Science, online erschienen am 07.06.2026.
Abstract (Originalsprache)
Machine learning models applied to heart disease prediction commonly report accuracies greater than 90% when applied to public benchmarks.The majority of these results, however, are obtained by training and testing on the same data.This paper addresses a more fundamental question: will these models still perform well when presented with patient records from an external source?We train five popular classifiers -logistic regression, k-nearest neighbors, a support vector machine, random forest, and XGBoost -on two separate sets of heart disease patient data, one consisting of 303 records from the UCI Cleveland set and the second composed of an amalgamation of the Cleveland, Hungarian, and Statlog cohorts amounting to 1190 records.On a single set, random forest is 92.9% accurate, with an AUC of 0.97 reflecting the same superior performance; however, when the same model is evaluated against the other data set, its accuracy drops to approximately 45% and the AUC to less than 0.55, that is, not significantly better than a random guess.The decline in performance is consistent across both classifier type and direction.Our finding suggests that high single-set accuracy has little bearing on a model's true clinical applicability and strongly advocates for the use of external validation and common data standards as an intrinsic part of any disease prediction study.
Zitieren
APA
(2026). ON THE GENERALIZATION GAP IN MACHINE LEARNING MODELS FOR HEART DISEASE PREDICTION: A CROSS-DATASET STUDY. International Research Journal of Modernization in Engineering Technology & Science. https://doi.org/10.56726/irjmets101546
Vancouver
ON THE GENERALIZATION GAP IN MACHINE LEARNING MODELS FOR HEART DISEASE PREDICTION: A CROSS-DATASET STUDY. International Research Journal of Modernization in Engineering Technology & Science. 2026. doi:10.56726/irjmets101546
BibTeX
@article{euvimed2026,
title = {ON THE GENERALIZATION GAP IN MACHINE LEARNING MODELS FOR HEART DISEASE PREDICTION: A CROSS-DATASET STUDY},
journal = {International Research Journal of Modernization in Engineering Technology & Science},
year = {2026},
doi = {10.56726/irjmets101546}
}
EUVIMED