A Hybrid Feature Extraction-Based Prediction System (HFEPS) for Early Diagnosis of Chronic Kidney Disease

The mortality rate of chronic kidney disease (CKD) was reported to be 1.2 million deaths worldwide due to existing challenges in diagnosis, such as required time, inadequate patient education, and poor communication. Therefore, an accurate prediction system for CKD is imperative to reduce the burden on the health care system. This research paper presents a hybrid feature extraction-based prediction system (HFEPS) for chronic kidney disease. This system can help in the early analysis of kidney disease based on the patient's record. The dataset of standard chronic disease available in the UCI (University of California, Irvine) machine learning repository was used for the recognition process. To develop the prediction system, the authors first applied the mean imputation method and one-hot encoding for preprocessing. Consequently, min-max scaling and SMOTE (Synthetic Minority Oversampling Technique) were applied. Furthermore, singular value principal component linear discriminant analysis (SVPC-LDA), a hybrid feature extraction method, is used for dimensionality reduction. Six high-performance classifiers are applied for classification: Gaussian NB, Linear SVC (Support Vector Classifier), Logistic Regression CV, Nu SVC, Passive Aggressive, and K-Neighbors. The combination of hybrid feature extraction and passive-aggressive algorithm attained the highest accuracy of 99%. In addition, the sensitivity of the passive-aggressive classifier improved by 4.26%, its specificity by 2.00%, its precision by 3.92%, and its AUC (Area Under the Curve) by 1.08%. Moreover, a feature reduction of 70.83% was achieved. The proposed HFEPS (Hybrid Feature Extraction Prediction System) can be employed for accurate and effective chronic kidney disease diagnosis. Extend the proposed technique to include more ailments such as heart disease, cancer, and diabetes.