Enhancing Diabetes Prediction With Minimal Processing Time Using Catboost: A Comparative Study.

Area of Science:

• Computational biology
• Medical informatics
• Machine learning applications

Background:

• Diabetes mellitus poses global health challenges, necessitating accurate prediction for early intervention.
• Machine learning (ML) offers advanced tools for improving predictive accuracy in healthcare.
• Ensemble ML methods are known to enhance predictive performance in complex datasets.

Purpose of the Study:

• To evaluate various ML classifiers, including ensemble methods, for diabetes prediction.
• To compare the computational efficiency and predictive performance of different ML models.
• To introduce and assess the CatBoost classifier for rapid and accurate diabetes risk identification.

Main Methods:

• Exploration of individual ML classifiers: decision trees, random forests, k-nearest neighbors, Naive Bayes, AdaBoost, XGBoost, and multilayer perceptron (MLP).
• Implementation and evaluation of ensemble configurations of these ML classifiers.
• Comparative analysis of model performance based on prediction accuracy and execution time, focusing on MLP, CatBoost, and ensemble models.

Main Results:

• The multilayer perceptron (MLP) was identified as the best-performing individual ML model.
• The proposed CatBoost classifier demonstrated superior computational efficiency, with an execution time of 4.27 seconds.
• CatBoost was approximately 98.64% faster than the ensemble model (314.96 seconds), highlighting its advantage in speed.

Conclusions:

• CatBoost offers a significant advantage in computational efficiency for diabetes prediction compared to ensemble methods.
• The study contributes to advancing precision medicine through efficient ML-based diabetes risk assessment.
• Leveraging diverse ML strengths enhances personalized healthcare strategies for individuals at risk of diabetes.