ARTIFICIAL INTELLIGENCE FRAMEWORK FOR ACCURATE DIABETES CLASSIFICATION USING ENSEMBLE LEARNING

Abstract

Diabetes mellitus is a rapidly increasing chronic metabolic disorder that presents considerable difficulties for worldwide healthcare systems due to its widespread occurrence, protracted diagnosis, and related consequences. Accurate early prediction of diabetes is essential for prompt treatment initiation and efficient disease control. When dealing with complicated clinical data, traditional diagnostic methods and single machine learning classifiers can have trouble with accuracy, overfitting, and generalisation. This article introduces an AI framework for precise diabetes classification with ensemble learning techniques to overcome existing constraints. The proposed architecture incorporates trimmed bagging, weighted model aggregation, and layered ensemble artificial neural networks to enhance prediction accuracy and reliability. A meta-learner combines the findings of several basic classifiers that were trained on different parts of the data to make better decisions. Pruning procedures are used to get rid of learners that aren't needed, make the process easier, and lower the chance of overfitting. Weighted voting also helps make accurate classification better. The model's performance is measured using accuracy, precision, recall, F1-score, and AUC on typical diabetic datasets. The experiment shows that the suggested ensemble architecture is more accurate, stable, and reliable than standalone neural network models and traditional machine learning. The method that was created is a useful clinical decision support tool for finding diabetes early and figuring out how likely someone is to have the disease. This leads to better healthcare analytics and better outcomes for patients.