Chronic kidney disease prediction using machine learning techniques

Abstract

In today’s fast-paced world, maintaining health often takes a backseat until visible symptoms arise. Unfortunately, certain diseases, like Chronic Kidney Disease (CKD), develop silently, presenting no noticeable symptoms in the early stages. This delay in detection often leads to severe complications, including kidney failure, cardiovascular disease, or even death. CKD’s silent progression highlights the critical need for proactive and predictive healthcare tools that can identify risks early. Machine Learning (ML) offers a promising solution, capable of analyzing vast amounts of data and predicting potential health risks with high accuracy. In this study, we explored the potential of nine ML techniques for predicting CKD: K-nearest Neighbors (KNN), support vector machines (SVM), logistic regression (LR), Naïve Bayes, Extra Tree Classifiers, AdaBoost, XG Boost, and Light GBM. Using a dataset obtained from Kaggle.com with 14 attributes and 400 records related to CKD, we aimed to identify the most effective model for this task. The attributes included clinical parameters such as blood pressure, specific gravity, albumin, sugar, and more, providing a comprehensive foundation for prediction. Each ML model was meticulously trained and tested, with hyperparameters fine-tuned to achieve optimal performance. Feature scaling and data preprocessing were conducted to ensure the models handled the dataset effectively. Evaluation metrics, including accuracy, precision, recall, F1-score, and ROC-AUC, were used to assess performance. Among the models, LightGBM emerged as the top performer, achieving an impressive accuracy of 99.00%. This model reformed its counterparts due to its ability to handle imbalanced datasets, fast training speed, and exceptional performance in capturing complex patterns.