A controlled ablation study of clinically motivated engineered features for heart disease prediction on the Statlog dataset

Heart disease prediction models increasingly combine machine learning with structured clinical knowledge and explainable artificial intelligence. However, performance claims on small benchmark datasets can be difficult to interpret when feature construction, data partitioning, and metric calculations are not fully reproducible. This study evaluates 13 clinical predictors and eight deterministic, clinically motivated indicators using the UCI Statlog (Heart) dataset (270 records). A fixed stratified 80:20 train–test split (seed 42) was used, with five-fold stratified cross-validation restricted to the 216-record training partition. The indicators were derived directly from the original predictors; the original OWL/SWRL ontology, terminology mappings, source code, and historical prediction artifacts were unavailable. The experiment therefore evaluates deterministic feature augmentation, not formal ontology reasoning. On the 54-record test set, the clinical-only DNN achieved 83.33% accuracy, 80.00% precision, 83.33% recall, 81.63% F1-score, and 0.867 ROC-AUC. The augmented DNN achieved 79.63% accuracy, 72.41% precision, 87.50% recall, 79.25% F1-score, and 0.879 ROC-AUC. These single-split differences are descriptive and should not be interpreted as evidence of improvement: all five cross-validation mean metrics favoured the clinical-only model. The original SHAP figure could not be validated against the reported input features and was removed; no quantitative SHAP findings are claimed in this revision. The results support cautious evaluation of deterministic feature engineering and do not establish clinical effectiveness or the independent value of ontology reasoning.

Authors

Institutions

Publication Details

Journal
Frontiers in Artificial Intelligence
Published
2026-11-20
DOI
https://doi.org/10.3389/frai.2026.2010348
Primary Topic
Artificial Intelligence in Healthcare
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A controlled ablation study of clinically motivated engineered features for heart disease prediction on the Statlog dataset

R. Vijayan, U. Priyadharshini
2 links
Frontiers in Artificial Intelligence
Artificial Intelligence in Healthcare
article

A controlled ablation study of clinically motivated engineered features for heart disease prediction on the Statlog dataset

R. Vijayan, U. Priyadharshini
article en

Abstract

Heart disease prediction models increasingly combine machine learning with structured clinical knowledge and explainable artificial intelligence. However, performance claims on small benchmark datasets can be difficult to interpret when feature construction, data partitioning, and metric calculations are not fully reproducible. This study evaluates 13 clinical predictors and eight deterministic, clinically motivated indicators using the UCI Statlog (Heart) dataset (270 records). A fixed stratified 80:20 train–test split (seed 42) was used, with five-fold stratified cross-validation restricted to the 216-record training partition. The indicators were derived directly from the original predictors; the original OWL/SWRL ontology, terminology mappings, source code, and historical prediction artifacts were unavailable. The experiment therefore evaluates deterministic feature augmentation, not formal ontology reasoning. On the 54-record test set, the clinical-only DNN achieved 83.33% accuracy, 80.00% precision, 83.33% recall, 81.63% F1-score, and 0.867 ROC-AUC. The augmented DNN achieved 79.63% accuracy, 72.41% precision, 87.50% recall, 79.25% F1-score, and 0.879 ROC-AUC. These single-split differences are descriptive and should not be interpreted as evidence of improvement: all five cross-validation mean metrics favoured the clinical-only model. The original SHAP figure could not be validated against the reported input features and was removed; no quantitative SHAP findings are claimed in this revision. The results support cautious evaluation of deterministic feature engineering and do not establish clinical effectiveness or the independent value of ontology reasoning.

Frontiers in Artificial IntelligenceVol. 9
Vellore Institute of Technology University (IN)
Good health and well-being
Artificial Intelligence in Healthcare
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon