Effectiveness and Hyperparameter Selection of Ensemble Models of Artificial Intelligence in Diagnosing Diabetes

Abstract

This study examined models for evaluating and optimizing the performance of machine learning and artificial intelligence algorithms in medical diagnostics, in particular, for solving the problem of multi-class classification of diabetes mellitus. In the study, for a diabetes dataset prepared from real clinical data, the target column "Diabetes" was divided into three classes: "healthy," "prediabetes," and "diabetes," and the dataset was divided into training sets and test sets. In this study, the performance of ten modern machine learning models, GBoost, KNN, Stacking MLP, MLP, Stacking RF, ExtraTrees, Stacking Bagging, XGBoost, LightGBM, and HistGradientBoosting, was tested using various evaluation metrics, including accuracy, precision, recall, and F1-score. Preliminary results showed high accuracy and generalization ability in the XGBoost models 0.9048 accuracy, ExtraTrees 0.9026 accuracy, LightGBM 0.9055 accuracy, and HistGradientBoosting 0.9066 accuracy, based on ensemble and gradient boosting. The results of these models after hyperparameter tuning in the next step were: HistGradientBoosting 0.9092 accuracy, LightGBM 0.9095 accuracy, ExtraTrees 0.9066 accuracy, and XGBoost 0.9063 accuracy. Precision, recall, and F1-score also improved accordingly. In this study, the optimal hyperparameters for the selected models were analyzed in tabular form. The results of the study show that the performance of ensemble-based and boosting-based models in multi-class classification was examined in the clinical dataset. The results of this study provide a scientific and practical basis for the use of artificial intelligence technologies in diabetes diagnosis and create a solid foundation for the development of modern approaches to the medical system of Uzbekistan.

Ähnliche Arbeiten