Early prediction of chronic kidney disease via ensemble-deep-learning approach and improved dimensionality reduction approach

This research proposes an ensemble-deep-learning approach and an enhanced dimensionality reduction method to predict chronic kidney disease (CKD) in its early stages. The proposed model has four major phases: pre-processing, feature extraction, feature selection, and CKD prediction. The pre-processed data is subjected to data cleaning and standardisation. Features are then extracted using PCA, statistical and higher-order statistical approaches, and the Spearman's rank correlation coefficient. Optimal features are selected using the hybrid optimisation model, which combines lemurs optimisation and harmony search algorithm. The CKD prediction phase uses an ensemble-deep-learning approach, including a transformer network, autoencoder, and feedforward neural network. The proposed model's accuracy, precision, F1 score, recall, FNR, FPR, and FDR are compared with existing models, showing its superiority. The proposed model aims to predict CKD in its early stages, potentially reducing its growth rate and associated issues.