Integrative Feature Selection and Enhanced Classification of Chronic Kidney Disease through Novel Weight Convolutional Neural Network Fusion

Abstract

Introduction: Chronic Kidney Disease (CKD) is a progressive condition marked by a progressive decline of renal function over time, resulting in complications like electrolyte imbalances and cardiovascular issue.

Objectives: Elucidating the complex ethology and global impact of CKD emphasizes the necessity for early detection and holistic management strategies to alleviate its far-reaching public health burden.

Methods: This study explores three primary feature selection techniques: Embedded, Wrapper, and Filter techniques in machine learning. Specifically, Recursive Feature Elimination with Cross-Validation (RFECV) is applied as the Wrapper method, aimed at developing a robust machine learning model by iteratively discarding irrelevant features. The research focuses on identifying and utilizing the most relevant features for training the model, enhancing its predictive performance by eliminating redundant or less informative attributes. This approach aims to optimize model accuracy and efficiency by iteratively selecting the most impactful features from the dataset. This research introduces the NWCNN as a novel classification method. The NWCNN approach is enhanced by integrating the LGWO technique, aiming to optimize both computation and classifier performance.

Results: The suggested model for the CKD dataset is planned for implementation in the Python tool. When compared to existing methods, the proposed model demonstrates a significantly higher accuracy. The proposed model for chronic kidney disease (CKD) classification achieved exceptional performance metrics, including an accuracy of 99.86%, precision of 98.70%, recall of 100%, F1 score of 99%, specificity of 100%, Matthews Correlation Coefficient (MCC) of 97.80%, and negative predictive value (NPV) of 99.50%.

Conclusions: The research explores the feature selection methods in machine learning, such as Filter, Wrapper, and Embedded methods, with emphasis on RFECV as the Wrapper method generally aims to improve the ability of ML models to predict CKD has been improved by selecting the most appropriate quality. The classification technique is further advanced with the introduction of the NWCNN and the incorporation of LGWO. By optimizing the computation and classification using LGWO, the NWCNN method seeks to increase efficiency and accuracy.