AI-powered chronic kidney disease detection integrating TabNet and LSTM for precision diagnosis

Abstract

Chronic kidney disease (CKD) is a leading cause of morbidity and mortality worldwide, with early detection being vital for effective treatment and management. Traditional diagnostic methods often struggle with the difficulty of CKD datasets, leading to delays in diagnosis and intervention. Despite advancements in machine learning, there remains a need for more effective and accurate models that can manage the intricacies of CKD data. This paper proposes an advanced AI-based system for early detection and diagnosis of CKD by developing an AI-based system that can efficiently handle complicated healthcare data. The workflow begins with data collection, where the CKD dataset, including patient health features, is gathered. The data is then preprocessed, involving missing data imputation using Expectation Maximization and outlier removal with One-Class SVM. Feature extraction is performed using Principal Component Analysis to reduce dimensionality while retaining significant features. The extracted features are passed through the TabNet model for feature selection and the long short-term memory (LSTM) for capturing temporal dependencies. Finally, the system makes abnormality detections, providing predictions based on the extracted features and temporal patterns. The model achieves a performance of 98.58% accuracy, 98.89% precision, 98.24% recall, and 98.56% F1-score, outperforming existing models such as XGBoost, SVM, and KNN. This work presents an effective combination of TabNet for feature selection and LSTM for sequential pattern recognition, resulting in a robust system for early CKD detection with minimal false positives and false negatives.