A Machine Learning Based Diabetes Mellitus Classification on Weighted Class using Inverse Class Frequency Method

A lot of research works exist on classifying Pima Indians diabetes mellitus classification. But many did not consider the imbalanced nature of the dataset and compared the classification algorithms on mainly accuracy. The accuracy metric is reliable only if the dataset is class-balanced. The imbalance dataset leads to biased model predictions, where the majority class dominates the learning process, and the minority class is often overlooked. The F1-Score and Receiver Operating Characteristic (ROC) curve become significant here. The F1-score is a comprehensive evaluation that provides a more robust evaluation metric, especially when the minority class is of particular interest. The ROC curve gives an overly optimistic assessment of the model's performance for imbalanced datasets. The proposed work uses Inverse Class Frequency method for assigning weights to prevalent classes in the dataset for improving F1-Score and reduces biasness. Six classification models, Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Gradient Boost (GB), Support Vector Machine (SVM), and Soft Voting Classifier (SVC), are investigated. The results are exhaustively compared on normalized original input features and PCA extracted features using the baseline equally weighted feature classes and the Inverse Class Frequency based weighted feature classes on 80:20 train-test split and 5-Fold cross validation. The experimental results show improved F1-Score and Area Under the ROC Curve (AUC) on weighted class and validate the stable and smooth results from the 5-Fold cross validation.