Early and accurate prediction of diabetics based on FCBF feature selection and SMOTE

Diabetes is a chronic hyperglycemic disorder. Every year hundreds of millions of people around the world have diabetes. The presence of irrelevant features and an imbalanced dataset are significant issues to train the model. The availability of patient medical records quantifies symptoms, body characteristics, and clinical laboratory test values that can be used in the study of biostatistics aimed at identifying patterns or characteristics that cannot be detected by current practice. This work proposes a machine learning-based healthcare model for accurate and early detection of diabetics. Five machine learning classifiers such as logistic regression, K-nearest neighbor, Naïve Bayes, random forest, and support vector machine are used. Fast correlation-based filter feature selection is used to remove the irrelevant features. The synthetic minority over-sampling technique is used to balance the imbalanced dataset. The model is evaluated with four performance measuring matrices: accuracy, sensitivity, specificity, and area under the curve (AUC). An experimental outcome shows few relevant features are needed to enhance the accuracy of the developed model. The RF classifier achieves the highest accuracy, sensitivity, specificity, and AUC of 97.81%, 99.32%, 98.86%, and 99.35%.