Analysis of the diabetes dataset using a SMOTE machine learning approach

Diabetes Mellitus affects huge number of people worldwide. Early identification is essential, if we were to limit this huge healthcare problem. The imbalance in class among datasets is a serious concern for machine learning solutions, particularly in generating biased predictions and poor performance for minority classes. The study investigates the effect of SMOTE on performance across multiple machine learning models on diabetes dataset. Before SMOTE, it was found that although the models score satisfactory overall accuracy , yet their average recall and AUC rates were low for the cases with a limited number of samples in the training datasets. In contrast, after applying SMOTE, the changes in recall, precision and the area under the curve noticeably improved, and most of the models met their high-performance classification criteria. There were substantial improvements, in particular, in ensemble and neural networks models. These experiments demonstrate that the synthetic minority over-sampling is efficient in balancing the datasets, thereby giving model robustness and equitable class performance. This study emphasizes the need for pre-processing techniques such as SMOTE for addressing imbalanced data problems in real-life situations.