Results of heterogeneous ensemble models on OpenML-CC18

Summary Results of different classifiers including heterogeneous models on the OpenML-CC18 curated tabular classification benchmarks. Data used for the ICPR2026 submitted paper Choosing a heterogeneous ensemble method in tabular classification by Vassili MAILLET, Jesús ANGULO and Pierre JOUVELOT. Specificities Ensemble methods: Stacking, Blending, Soft Voting, Hard Voting, Bagging, UMAP Dynamic Classifier Selection and Subset Classifier Selection. Base learner types: Decision Tree, SVM, Logistic Regression, Random Fores, XGBoost, Gaussian Naïve Bayes, Most Common (custom). Training: Five repetitions of 10-fold stratified cross validation Metrics: ROC AUC OVO, Accuracy, Matthew's Correlation Coefficient, Confusion Matrix Computer: CPU: AMD Ryzen 7 5800h RAM: 16 Gb of memory Software: Python 3.10 or 3.12 depending on cases scikit-learn 1.5 or 1.7 depending on cases Note Instructions on how to use them and more specificities in the README. The .zip contains all JSON files as well as the README.