Generative AI and machine learning framework for early detection of chronic kidney disease

Abstract

Chronic Kidney Disease (CKD) is a serious public health problem which is affecting nearly one in 10 adults all over the world. However, early detection is challenging because the clinical signs are subtle and the low sensitivity of detection tests. This research suggests a new approach that combines Machine Learning (ML) and Generative AI to boost the performance of the model and to solve the problem of data scarcity. Synthetic data were created using Conditional Tabular Generative Adversarial Network (CTGAN) to supplement the publicly available UCI CKD dataset. This helps to fix the problem faced like class imbalance and to increase data diversity while maintaining clinical validity. The original UCI CKD dataset had 400 records and 24 attributes. Two samples were removed due to missing values. It was extended with 500 synthetic data generated by CTGAN which resulted in balanced dataset of 898 records which was also approved by a certified urologist. 3 ML models like eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM) and Random Forest (RF) was used to train and evaluate the model using balanced dataset of 898 records. RF showed the best performance of 96.76 % accuracy, 0.98 precision, and a weighted F1-score of 0.97. This model outperformed on XGBoost and SVM models. This results shows that using generative data augmentation improves model generalization and diagnostic reliability, especially in situation where there is medical data scarcity.