Chronic kidney disease prediction using improved deep belief network with local search nearest neighbour optimization

Abstract

Chronic Kidney Disease (CKD) is one of the risky diseases that can threaten human life. Thus, automatic system is recommended for the early prediction. Hence, Artifical Intelligence (AI) based deep learning approach is designed in this current research for CKD prediction. However, attributes which is highly correlated to kidney diseases must be considered for effective disease prediction. In this pape a Local Search With Nearest Neighbour (LSNN) optimisation is proposed to select the most relevant attributes for training the deep net model. In the proposed LSNN optimisation, a K-fold cross-validation is applied to calculate the mean square error, which acts as the fitness function to select the best attributes using local search optimisation. Subsequently, the selected attributes are utilised for training the proposed Improved Deep Belief Network (IDBN). In IDBN, the performance of a conventional Deep Belief Network (DBN) is improved by utilising a hybrid atom crow optimisation in the training phase. The analysis is conducted on different scenarios, and the selected 15 attributes are better, with an accuracy of 98% and an error is 2%. Ultimately it proves that the proposed deep learning with LSNN-based attribution is better for the early prediction of CKD.