Abstract

Chronic Kidney Disease (CKD) is a progressive disorder that gradually impairs kidney function and often remains undetected until advanced stages, making early prediction essential for timely clinical intervention. Conventional machine learning methods generally struggle to model the temporal dependencies present in longitudinal patient records, resulting in limited prediction performance. This study proposes a Long Short-Term Memory (LSTM)-based deep sequential framework for predicting CKD progression using multivariate time-series clinical data. The proposed model utilizes longitudinal features, including estimated glomerular filtration rate (eGFR), serum creatinine, blood pressure, glucose, albumin, and hemoglobin, to learn complex temporal patterns associated with kidney function decline. Data preprocessing involves missing-value imputation, normalization, outlier removal, temporal feature engineering, and sliding-window sequence generation. To improve model robustness and generalization, dropout and L2 regularization are incorporated together with the Adam optimizer and Huber loss. Experimental results demonstrate that the proposed model achieves a Mean Absolute Error (MAE) of 3.08 mL/min/1.73 m², a Root Mean Square Error (RMSE) of 4.11 mL/min/1.73 m², a Mean Absolute Percentage Error (MAPE) of 6.35%, and a coefficient of determination (R²) of 0.93. Comparative analysis with Linear Regression, Random Forest, and Support Vector Machine demonstrates the superior predictive capability of the proposed framework. These findings indicate that the proposed LSTM model provides an accurate and reliable solution for intelligent CKD progression prediction and can effectively support early diagnosis, personalized treatment planning, and clinical decision-making in modern healthcare systems.

References

• J. L. Hill et al. 2016. "Global prevalence of chronic kidney disease – A systematic review and meta-analysis." PLOS One 11, 7 (2016).
• GBD Chronic Kidney Disease Collaboration. 2025. "Global, regional, and national burden of chronic kidney disease, 1990–2023." The Lancet (2025).
• H. Wang et al. 2025. "Global burden of chronic kidney disease and its risk factors." Frontiers in Public Health 13 (2025).
• GBD 2021 Chronic Kidney Disease Collaborators. 2021. "Global burden of chronic kidney disease: A systematic analysis." The Lancet (2021).
• National Kidney Foundation. 2024. "About Chronic Kidney Disease."
• D. Dua and C. Graff. 2019. UCI Machine Learning Repository. University of California, Irvine.
• S. Hochreiter and J. Schmidhuber. 1997. "Long short-term memory." Neural Computation 9, 8 (1997), 1735–1780.
• Z. C. Lipton, D. Kale, and R. Wetzel. 2015. "Learning to diagnose with LSTM recurrent neural networks." arXiv preprint arXiv:1511.03677 (2015).
• Y. LeCun, Y. Bengio, and G. Hinton. 2015. "Deep learning." Nature 521 (2015), 436–444.
• National Kidney Foundation. 2024. "Estimated Glomerular Filtration Rate (eGFR)."
• B. A. Goldstein et al. 2015. "Opportunities and challenges in developing risk prediction models with electronic health records data." Journal of Biomedical Informatics 58 (2015), 113–121.
• S. M. Lundberg and S.-I. Lee. 2017. "A unified approach to interpreting model predictions." In Advances in Neural Information Processing Systems (NeurIPS).
• World Health Organization. 2024. "Chronic kidney disease."
• Centers for Disease Control and Prevention. 2024. "Chronic Kidney Disease Basics."
• J. Esteva et al. 2019. "A guide to deep learning in healthcare." Nature Medicine 25 (2019), 24–29.