An Intelligent Diabetes Prediction System Augmenting Feature Selection and Balancing Techniques

Diabetes seems to be a prevalent disease nowadays across the whole globe and is found to be very common on every doorstep. The consequences of this disease are very severe if it remains unnoticed during its initial stage. It affects populations of every age group. The causes of this disease are attributed to genetic links, lack of physical exercise, unhealthy food habits, overstress, etc. Many recent machine learning models have been proposed to predict this disease by considering the historical data of patients who have diabetes. This work attempts to choose a robust and reliable machine learning model for diagnosing the disease from seven widely used classification algorithms such as; Logistic Regression, Decision Tree, Naïve Bayes, Support Vector Machine, Random Forest, XGBoost, and AdaBoost. The model is validated with the PIMA diabetes dataset. The efficiency of the models is augmented by addressing the imbalanced characteristics of the dataset by applying SMOTE technique and a univariate feature selection technique, ‘chi-square’ for feature selection. The experimental findings confirm that the Random Forest model has outperformed other models in terms of accuracy, precision, recall, F-score, and AUC.