Investigating Anomalies in Diabetes Time-Series Data Using Boosting Algorithms Family : A Comparative Study of Ensemble Approaches

Diabetes mellitus, a chronic metabolic disorder characterized by elevated blood glucose levels, poses significant challenges in achieving optimal glycemic control, particularly through traditional treatment methods. The complexities of managing diabetes are further heightened when dealing with multivariate time-series data that encompass various parameters such as blood glucose levels, insulin dosages, and lifestyle factors. This study explores the application of boosting algorithms for anomaly detection in such data, focusing on AdaBoost, XGBoost, LightGBM, Gradient Boosting, and CatBoost. Accurate detection of anomalies is crucial for enhancing diabetes management and improving patient outcomes. While traditional machine learning methods like K-Nearest Neighbors and One-Class SVM are commonly used, they often struggle with high-dimensional and multivariate data. Boosting algorithms, designed to improve model performance, offer a promising solution. This research evaluates these algorithms in terms of their predictive accuracy and ability to handle complex multivariate time-series data. Results show that LightGBM outperforms other models, demonstrating the highest R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value and lowest Mean Squared Error, making it more effective for datasets with significant variability and multiple influencing factors.