CAD-CKD: a computer aided diagnosis system for chronic kidney disease using automated BiGSqENet in the Internet of Things platform

Abstract

A fast and efficient disease diagnosis system based on IoT data is challenging since the data has high data redundancy. To formulate this, many pre-processing techniques and feature-based classification techniques are used. However, the processing time and the computational effort of these processes are high. To counteract this, this article introduces a novel outlier removal model that uses a three-step clustering process. Instead of using distance- or density-based clustering, a directional density ratio is evaluated here, which provides the direction, distance, and density of the samples. Based on the directional density ratio, samples are clustered using K-Means, Reverse K-Nearest, and agglomerative clustering methods. The clustered features are provided as input to a novel deep neural network that hybridizes bidirectional GRU and squeeze-and-excitation residual networks. To improve the accuracy of the prediction, a Multiverse Jackal optimization (MJO) is used, which reduces the cross-entropy of the proposed classifier. As a result, the proposed method has resulted in an accuracy of 99.8% with an execution time of 103.7 ms, which is 33.6% faster than existing state-of-the-art methods.