Traditional Methods and Modern Approaches Based on Ensemble Algorithms for Decision-Making in Diagnostics Using Medical Data
Аннотация
Currently, artificial intelligence and machine learning algorithms are being applied in all fields. This study compares existing methods and approaches with machine learning algorithms, highlighting their advantages and disadvantages. This article provides an in-depth analysis of the practical capabilities of modern machine learning and artificial intelligence algorithms in the early diagnosis and prediction of diabetes mellitus. In this research, experimental results were obtained for multi-class classification using ensemble models HistGradientBoosting and LightGBM, based on patients' clinical and demographic indicators. These models demonstrated the ability to distinguish between healthy, pre-diabetic, and diabetic categories with high accuracy, as well as to create personalized risk profiles for each patient. The experimental results confirmed that the selected models achieved accuracy and stability above 0.90 according to the main criteria for assessing classification quality: accuracy, recall, precision, F1-score, and confusion matrix results. Additionally, the modern stages of the data-driven decision-making process in medical diagnostics were scientifically described, including data collection, cleaning, feature selection, optimal model selection, and interpretative visualization of results. The study specifically highlighted the crucial importance of explainable AI technologies in enhancing the transparency and reliability of clinical decisions. Furthermore, the article emphasized the importance of balanced quality in medical data. The obtained results indicate that integrating modern ML and AI algorithms can significantly improve the efficiency of diagnosing diabetes and making clinical decisions, potentially yielding better outcomes than existing methods in terms of both time and accuracy.