A Novel Hybrid Model for Efficient Prediction of Diabetes Using Machine Learning

Diabetes mellitus, a long-term metabolic disease marked by elevated blood glucose levels, has a substantial negative impact on public health worldwide. Accurate prediction is essential for effective diabetes treatment due to its increasing prevalence. Machine Learning (ML) addresses the issues of this chronic metabolic condition, which provides a promising answer by enabling individualized therapies. This paper presents a comprehensive study on diabetes prediction using ML techniques, focusing on a hybrid approach stacked ensemble models. The research employs the Indian Pima dataset from Kaggle. The primary objective is to assess various ML models with and without preprocessing techniques. Preprocessing involves data cleaning, normalization, and feature engineering to enhance predictive accuracy. Overall, the results underscore the importance of preprocessing in improving predictive performance and highlight the efficacy of hybrid ensemble techniques in diabetic prediction tasks. Moreover, the hybrid approach utilizes a stacked ensemble comprising Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP) as base models and Logistic Regression as a meta-model. The hybrid approach achieves superior performance compared to individual models with 95.1% accuracy. This methodology effectively combines the strengths of diverse algorithms, mitigating their weaknesses and yielding more robust predictions. It provides substantial benefits by integrating ML algorithms, statistical techniques, and expert knowledge.