CKD-LSTM: Chronic kidney disease detection using LSTM integrated with oversampling technique

Abstract

Chronic Kidney Disease (CKD) is a global health concern because it has a high fatality rate. Most of the patients come to know about CKD when it becomes fatal as in the early stages of CKD no symptoms are shown. Early diagnosis of CKD aids patients in getting appropriate treatment timely and saves their lives. Deep learning models play a vital role in achieving effective and fastest diagnosis. In this study, we propose a LSTM model to detect CKD using the UCI dataset consisting of 400 samples. In the research, to deal with the data imbalance problem SMOTE technique is used and to fill up the large number of null values, random value imputation method is performed. Along with 5-fold cross-validation, in result analysis, it found that the proposed LSTM model obtained an accuracy of 97.2% which outperforms the state-of-the-art. Besides accuracy, the proposed model shows significant improvements in precision (95.8%), recall (98.8%), F1 score (97.3%), and AUC (97.2%).