Abstract:
One best example of chronic diseases and one of the highest causes of morbidity is diabetes whose early detection can help avert a catastrophic health risk. Early and precise detection of diabetes can lead to improved patient outcomes and facilitate health care providers with preventive care. In this paper Proposed Hybrid Ensemble Model (stacking ensemble meta-learner model for feature fried formulated on the basis of 2,768 patient records which contain 10 attributes related to some clinical characteristic are used for diabetes classification namely Pregnancies, In the pre-processing stage, missing values were imputed with the number of median and mode, imbalance between classes were handled by SMOTE and features were normalized. EDA through histograms, correlation heatmaps and distribution plots was carried out to preliminarily inspect co-feature interactions and how they affect diabetes end-results. The Proposed Hybrid Ensemble Model was constructed using the stacking of base learners such as Decision Tree and Random Forest, and a meta learner named Logistic Regression. The model was built by splitting the images into 80% train set and 20% test set and validated with 5-fold crossvalidation the above experimental results show that the Proposed model is better than Traditional Machine Learning in performance, and the., The accuracy, precision, recall, F1score are 99.72%, 1.0000, 0.9945, 0.9972, the AUCs are 1.0000. Experiments demonstrate that Proposed Hybrid Ensemble Model has the best generalization capability and robustness compared with the baseline model’s LR, DT, and RF diabetes prediction. This study suggests that the Proposed Hybrid Ensemble Model stacking ensemble model is an efficient and credible early detection tool for diabetes, which could assist clinical decision making.
