A Novel Approach Using Artificial Intelligence for Early Prognosis of Chronic Kidney Disease in Asian Population

Abstract

Purpose: The goal of this research project is to address the increasing prevalence of chronic kidney disease (CKD) and the challenges associated with its early detection. The study proposes a new approach that combines Linear Discriminant Analysis (LDA) to reduce the number of features and Artificial Neural Networks (ANN) for accurate and early identification of CKD. The primary objective is to create a smart decision-making system that can help nephrologists in India diagnose CKD in the early stages.

Method: The study is based on a dataset collected from DY Patil Hospital in Navi Mumbai. It consists of around 500 records with 21 attributes. To improve accuracy and reduce prediction time, the proposed model combines LDA and ANN. Feature selection is carried out using Recursive Feature Elimination (RFE), which identifies Creatinine, BUN, and Urea as crucial factors. The methodology includes preprocessing, hyperparameter tuning, and classification. Statistical analyses, such as hyperparameter values, Friedman's Test, and parallel computing evaluation, are used in the comprehensive methodology.

Results: The proposed hybrid model, referred to as a Hybridized LDA with ANN (HLDANN), outperforms traditional classifiers like SVM, LR, RF, DT, KNN, and even a standalone ANN. The model achieves an accuracy of 93.22% on a real-time dataset, surpassing other algorithms. Precision, recall, and F1 score metrics further validate the effectiveness of HLDANN. Parallel computing analysis demonstrates reduced prediction time with an increase in worker nodes.

Conclusion: Based on the study, the hybrid model is an effective method for detecting CKD at an early stage. It provides significant accuracy improvements compared to existing methodologies. The key features identified in the study align with known biomarkers, confirming the model's reliability. To further enhance the model's robustness, future research could explore additional biomarkers, diverse data sets, and non-invasive techniques. The proposed algorithm shows promise as a valuable tool for the medical community that could contribute to early CKD diagnosis and improved patient outcomes.