Optimizing Type II Diabetes Prediction Through Hybrid Big Data Analytics and H-SMOTE Tree Methodology

Abstract

In the last few years, Type II diabetes has become much more common worldwide, presenting major problems for both healthcare systems and individuals. Utilizing big data analytics has shown potential as a means of forecasting and managing persistent illnesses, like Type II diabetes. This paper proposes a novel hybrid approach that combines big data analytics techniques with an H-SMOTE tree algorithm for the prediction of Type II diabetes. The suggested method addresses the problems of class imbalance present in medical datasets and improves prediction accuracy by combining steps of feature selection, data preprocessing, and classification. In order to prepare raw data for analysis, it must first be cleaned, standardised, and transformed. Then, feature selection techniques are used to identify the most important factors that help predict Type II diabetes. This approach streamlines the predictive model and lowers its dimensionality. In the classification phase, an algorithm called the H-SMOTE tree is used. This method combines two existing techniques: the Hoeffding Adaptive Tree (HAT) and Synthetic Minority Oversampling Technique (SMOTE). The H-SMOTE tree tackles imbalanced data by creating synthetic samples for the under-represented class, while also adapting the decision tree structure as it receives new data.   Experiments show that this approach is effective in accurately predicting Type II diabetes. The researchers found that the H-SMOTE tree model outperformed other machine learning methods, both classic and recent ones. In other words, it was more accurate in predicting T2DM cases. This was evident in terms of several metrics, including how well it identified true positives (sensitivity), how well it avoided false positives (specificity), and its overall performance captured by the AUC-ROC score. Additionally, the proposed method displays resilience and scalability, rendering it apt for managing extensive medical datasets frequently encountered within healthcare domains.