Neural Nephroinformatics: Ensemble Strategies in Deep Learning for CKD Detection

Abstract

A burden on global health, chronic kidney disease (CKD) affects about 10% of adult population worldwide. It is acknowledged as one among the top 20 global causes of death. Although there exists treatment for chronic kidney disease, early detection of the illness can help mitigate the damage and slow down the disease's progression. Consequently, to improve the accuracy and effectiveness of the conventional chronic kidney disease diagnosis system, contemporary computer-aided approaches must be used. In the suggested ensemble model, Support Vector Machine (SVM) used as the stacking ensemble model, which mixes two hybrid deep learning models.  The dataset of CSV files used to detect CKD was obtained from the Kaggle repository in order to validate our model. Convolutional neural network- Gated Recurrent Unit (CNN-GRU) and Convolutional neural network-Long short-term memory (CNN-LSTM) are the two hybrid models employed in the model. With a high accuracy of 98.95%, the model produced generally accurate predictions. Recall and precision ratings of 98.56 and 100 respectively, show how accurate the classifications model can be.  The suggested stacking ensemble model was contrasted with both our own implementation and alternative methods for detecting CKD. The proposed approach outperforms other existing techniques in terms of performance, while utilizing the model to prevent overfitting.