An Innovative Ensemble Framework Using Decision Trees for Improved Early Prediction of Type 2 Diabetes

Resumen

The increasing prevalence of type 2 diabetes mellitus (T2DM) has led to an urgent need for predictive models capable of early detection and timely management. Key factors in forecasting T2DM risks are commonly characterized by their intricate connections between biological markers — features of these existing frameworks, such as the boosting ensemble, XGBoost and AdaBoost, have faced difficulty in tackling accurately. This research introduces an innovative decision tree-based T2DM detection and classification ensemble model (DDCEM). It combines the Pima Indians Diabetes Dataset (PIDD) and the Early-Stage Diabetes Risk Prediction Dataset (ESDRPD), in which decision trees (DT) with a random forest (RF) classifier are applied to achieve better predictive performance. Compared with the other traditional frameworks, the DDCEM model achieved an accuracy of 90%, which was appropriate and less accurate than the proposed methods, namely, the boosting ensemble (85%), XGBoost (82%) and AdaBoost (86%) methods. It was even better in precision (0.85), recall (0.80) and F1 score (0.82), demonstrating that these alterations efficiently minimized misclassified instances. Moreover, the model had a rapid response with low latency (0.15 s), suggesting its suitability for real-time healthcare settings. This paper effectively illustrates the establishment and validation of the DDCEM model, highlighting its efficiency in identifying and managing T2DM at an early stage. The proposed DDCEM model represents a notable advancement in T2DM prediction, but was not specifically introduced to replace decision trees or become a new addition to the random forest classifier techniques. There is potential for future work that involves leveraging different datasets across populations, applying advanced machine learning (ML) methodologies to account for the identified data imbalance issue and expanding our approach in predicting other related health conditions. This approach can improve the reliability and adaptability of models, leading to better healthcare results for patients.