Target leakage and the use of diagnostic variables in diabetes prediction models
Melike Tombaz, Nico Pfeifer, Sabrina Ehnert, Andreas Klaus Nussler · Nutrition and Diabetes · 2025
We read the article “Optimization of diabetes prediction methods based on combinatorial balancing algorithm” by Shao et al., 2024, with great interest [ 1 ]. The authors tackle a significant technical issue in biomedical data science: class imbalance in diabetes prediction models. By combining SMOTE with random under-sampling (RUS) and using Optuna for hyperparameter tuning of LightGBM, they enhance the ongoing efforts to boost model performance in imbalanced datasets.