Optimizing Type 2 Diabetes Classification with Feature Selection and Class Balancing in Machine Learning

Type 2 Diabetes (T2DM) is a crucial factor in patient survival and treatment effectiveness. Errors in diabetes detection lead to disease severity, high costs, prolonged healing time, and a decline in service quality. Additionally, a major challenge in developing Machine Learning (ML)-based detection decision support systems is the class imbalance in medical data as well as the high feature dimensionality that can affect the accuracy and efficiency of the model. This research proposes an approach based on feature selection (FS) and handling class imbalance to improve performance in type 2 diabetes. Several feature selection techniques such as Information Gain (IG), Gain Ratio (GR), Gini Decrease (GD), Chi-Square (CS), Relief-F, and FCBF can perform feature selection based on weighting ranking. Furthermore, to address the imbalanced class distribution, we utilize the Synthetic Minority Over-Sampling Technique (SMOTE). ML classification models such as Support Vector Machine (SVM), Gradient Boosting (GB), Tree, Neural Network (NN), Random Forest (RF), and AdaBoost were tested and evaluated based on the confusion matrix including accuracy, precision, recall, and time. The experimental results show that the combination of strategies for handling imbalanced classes significantly improves the predictive performance of ML algorithms. In addition, we found that the combination of feature selection techniques IG+AdaBoost consistently demonstrates optimal performance. This study emphasizes the importance of data preprocessing and the selection of the right algorithms in the development of machine learning-based T2DM detection systems. Accurate detection can reduce the severity of disease, lower treatment costs, speed up the healing process, and improve healthcare services.