Deep Learning-Powered Integrated Clinical Decision Support System for Chronic Kidney Disease Prediction

Abstract

This research work proposes a high level of machine learning to early and accurate prediction of Chronic Kidney Disease (CKD) via a hybrid deep learning pipeline evaluated on Chronic Kidney Disease Dataset. The initial stage preprocessing of the CKD dataset involves the use of Synthetic Minority Oversampling Technique (SMOTE), Adaptive Synthetic Sampling (ADASYN), and Random Under-sampling to balance and represent the distribution of CKD and non-CKD cases in the dataset. In an attempt to improve the quality of the features, the Principal Component Analysis (PCA) is used to reduce the dimensionality, which, in effect, captures the most discriminating components whilst reducing the amount of noise and redundancy. Simultaneously, Long Short-Term Memory (LSTM) networks are integrated to acquire both temporal and nonlinear relationships between clinical characteristics, which can be well-feature extracted in comparison with traditional models. The combined PCA-LSTM features are then inputted into a Deep Belief Network (DBN) classifier which takes advantage of the multilayer generative architecture to enhance the accuracy of CKD prediction and generalization. Experimental analysis shows that the proposed model has better predictive performance, and the accuracy, precision, recall, and F1-score are 97.8%, 96.5%, 98.1%, and 97.3% respectively, which is better than baseline machine learning models. The results confirm the appropriateness of using a hybrid approach to feature engineering and deep learning as a hybrid in clinical risk assessment. The given framework offers a stable and scalable solution to automated CKD diagnosis, which can help healthcare professionals to detect and intervene in time.