Context dependent bidirectional deep learning and Bayesian gaussian auto-encoder for prediction of kidney disease

Abstract

Chronic kidney disease (CKD) has emerged as a significant global health issue, leading to millions of premature deaths annually. Early prediction of CKD is crucial for timely diagnosis and preventive measures. While various deep learning (DL) methods have been introduced for CKD prediction, achieving robust quantification results remains challenging. To address this, we propose the context-dependent bi-directional DL and Bayesian gaussian autoencoder (CDBDP-BGA) method for CKD prediction. This approach utilizes clinical parameters and symptoms from a structured dataset. By incorporating context dependence into the bi-directional long short-term memory (Bi-LSTM) model, CDBDP-BGA efficiently redistributes the representation of information, enhancing its modeling capabilities. Feature selection is optimized using a BGA-based algorithm, which employs the Bayesian gaussian function. The SoftMax activation function classifies CKD into five distinct stages based on estimated-glomerular filtration-rate (eGFR), considering both symptoms (texture and numerical features) and clinical parameters (age, sex, and creatinine). Simulation results using two datasets demonstrate that CDBDP-BGA outperforms conventional methods, achieving 97.4% accuracy without eGFR and 98.7% with eGFR.