Efficient Chronic Kidney Disease Prediction: A Comparative Analysis Using Feature Selection and Machine Learning Models

Chronic nephropathy is considered one of the primary global causes of mortality. Blood pressure and diabetes are the main causes of this disease. Chronic kidney disease (CKD) can be prevented by early diagnosis, which can avoid other more serious complications such as end-stage renal disease (IRT). The early prediction of CKD is a very complex problem that requires highly developed and sophisticated solutions, such as artificial intelligence (AI) especially machine learning (ML) and deep learning (DL). This can also help improve healthcare for patients with chronic kidney disease by allowing for early detection of the disease. Our work aims to propose an artificial intelligence (AI) based approach to find the best classifier or algorithm for accurate and efficient prediction of chronic kidney disease. For our proposed approach to be effective, we conduct a thorough analysis and efficient prediction of chronic kidney disease. To achieve this, we employed effective and precise data preprocessing techniques. We utilized three machine learning algorithms: Random Forest (RF), AdaBoost, and Bagging, in addition to a hybrid model (Ada_Bag). For the training of these models, three feature selection methods were incorporated: LASSO, RELIEF, and Recursive Feature Elimination (RFE). Subsequently, we compared the performance of the machine learning models across these three feature selection techniques to identify the best method for each trained model, For the Lasso technique, the Random Forest (RF) model achieved a highest accuracy of 100% and using the RFE feature selection, the hybrid Ada_Bag model achieved an accuracy = 99,1% with 10 features.