Data Augmentation Approach for Type-2 Diabetes Prediction and Classification

Objectives: The main objective of this study is to develop a generalized machine learning framework for Type 2 diabetes prediction that integrates robust preprocessing, feature analysis, and advanced data augmentation techniques to enhance prediction accuracy. Methods: The proposed model of Type-2 Diabetes Prediction and Classification uses publicly available PIMA Diabetes Dataset includes missing value analysis, outliers detection, standardization as a pre-processing task, and a diabetic and non-diabetic class-wise data augmentation module using Adversarial Variational Auto-Encoder (AVAE) with Random Forest Classifier. The experimental results are evaluated and compared with Logistic Regression, Decision Tree, AdaBoost, SVC, Random Forest, GradientBoosting, and KNeighbors with cross-validation scores using mean accuracy. Findings: The experimental findings using RandomForestClassifier suggest a Type 2 diabetes prediction based on class-wise data augmentation has an accuracy level of 93.09%. Novelty: For medical dataset, class-wise data augmentation using Adversarial Variational Autoencoder plays crucial role while developing an accurate model. Keywords: Type 2 Diabetes, Data Augmentation, AVAE, Random Forest Classifier, Cross Validation Scores