Early prediction of chronic kidney disease based on ensemble of deep learning models and optimizers

Abstract

Recent studies have proven that data analytics may assist in predicting events before they occur, which may impact the outcome of current situations. In the medical sector, it has been utilized for predicting the likelihood of getting a health condition such as chronic kidney disease (CKD). This paper aims at developing a CKD prediction framework, which forecasts CKD occurrence over a specific time using deep learning and deep ensemble learning approaches. While a great deal of research focuses on disease detection, few studies contribute to disease prediction before it may occur. However, the performance of previous work was not competitive. This paper tackles the under-explored area of early CKD prediction through a high-performing deep learning and ensemble framework. We bridge the gap between existing detection methods and preventive interventions by: developing and comparing deep learning models like CNN, LSTM, and LSTM-BLSTM for 6–12 month CKD prediction; addressing data imbalance, feature selection, and optimizer optimization; and building an ensemble model combining the best individual models (CNN-Adamax, LSTM-Adam, and LSTM-BLSTM-Adamax). Our framework achieves significantly higher accuracy (98% and 97% for 6 and 12 months) than previous work, paving the way for earlier diagnosis and improved patient outcomes.