A Proposed Multilayer Perceptron Model and Kernel Principal Analysis Component for the Prediction of Chronic Kidney Disease

unfortunately, this stage is mostly detected at a late stage, leading to dialysis or transplantation. Early detection is important for the effective management of CKD. ML has shown success in the early prediction of CKD by using an algorithm that learns and predicts without being programmed. ML requires appropriate datasets for this process, and one of the aspects is dimensionality reduction, which addresses the challenges of unnecessary tests, high-cost tests and the use of redundant tests. Principal Component Analysis (PCA) is a widely used method for dimensionality reduction; however, it relies on linear transformation to identify relationships within features. Medical datasets such as CKD exhibit complex nonlinear features, which is important for exploring alternative dimensionality reduction methods that can rely on nonlinear transformation. This study aims to propose an ML approach that utilises kernel PCA to reduce dimensionality based on nonlinearity structures and enhance the prediction of CKD. We evaluated seven ML models on the different kernel functions of PCA. The ML models included random forest (RF), decision tree (DT), multilayer perceptron (MLP), support vector machine (SVM), extreme gradient boosting (XgBoost), adaptive boosting (AdaBoost), logistic regression (LR), and gradient boosting. The kernel functions used for dimensionality reduction are cosine principal component analysis (CPCA), polynomial principal component analysis (PPCA), radial basis principal component analysis (RPCA), sigmoid principal component analysis (SPCA) and linear principal component analysis (LPCA). The results of the study revealed that the MLP with RPCA, SPCA and CPCA achieved good performance in predicting CKD, with an accuracy score of 99% on DB1, and that the MLP with RPCA and SPCA achieved good performance in predicting CKD, with an accuracy score of 100% on DB2. The study showed how kernel PCA, which effectively reduces high dimensionality-based nonlinearity relationships, can positively affect the performance of predictive models and the power of dimensionality reduction toward disease prediction.