An Effective Diabetes Prediction System Using Machine Learning Techniques

Diabetes, a universal chronic disease, arises due to the lack of responsiveness to insulin. As the permanent cure for diabetes is not possible, so early detection and taking necessary treatments are the only way to mitigate the menace of other fatal diseases. Numerous researches have been conducted using different machine learning techniques for the early detection of diabetic patients. However, because of having missing values, irrelevant features, and imbalanced class distribution in the dataset, improving the prediction accuracy is always an arduous task. In this research, we present a Tree-Based machine learning model for classifying the Pima Indians Diabetes Dataset (PIDD). Mutual Information (MI) based feature selection technique is adopted to remove the less important features for better prediction. Finally, to improve the performance of Tree-Based algorithms, the Adaptive Boosting (AB) algorithm is employed. Upon comparison with experimental data, the Extra Tree (ET) algorithm as a base estimator of the AdaBoost classifier yields the highest accuracy of 90.5%. Therefore, our proposed Tree-Based machine learning model may assist healthcare practitioners for the diagnosis of diabetes.