Modelling of IDBN with LSNN based optimal feature selection for the prediction of CKD using real time data

Chronic Kidney Disease (CKD) is a critical condition induced by either reduced kidney functions or kidney pathology. The early diagnosis of CKD is considered significant to prevent the patient from a serious condition. In literature, various techniques are presented to detect CKD at the initial stage, but providing a better performance is still challenging, which causes the patients may not identify their diseases at the starting stage. Some of the noticeable drawbacks of existing techniques are higher-dimensional attributes, computational issues and high execution time. In order to overcome these concerns, this proposed paper focuses on designing an improved deep learning approach with optimal feature selection for the effective detection of CKD. The first stage of the proposed method is preprocessing, which performs digitalization, normalization and data filling. Then, a novel Local Search with Nearest Neighbour (LSNN) optimization is introduced for selecting effective features. Finally, the Improved Deep Belief Network (IDBN) is built using the obtained optimal features. The proposed CKD prediction model is validated using Benchmark and Real time datasets. The performance of the model is analyzed by considering five different cases. Each case has a different combination of attributes, and the performances are compared with conventional CKD prediction techniques like Deep Belief Network (DBN), Deep Neural Network (DNN), Multi-layer Perceptron (MLP), and Support Vector Machine (SVM). Both the prediction strategies provided better performance while using the optimally selected 15 features by LSNN. The proposed IDBN provided a maximum of 98% and 97% of accuracy, 0.01 and 0.02 of error value, 99% and 98% of precision, 98% and 99% of recall for benchmark and real time datasets, respectively. Ultimately the proposed IDBM with LSNN based techniques provided better performance for the prediction of CKD.