An innovative approach for predictive modeling and staging of chronic kidney disease

Diagnosing silent diseases such as chronic kidney disease (CKD) at an early stage is challenging due to the absence of symptoms, making early detection crucial to slowing disease progression. This study addresses this challenge by introducing a novel feature, the estimated glomerular filtration rate (eGFR), calculated using the modification of diet in renal disease (MDRD) formula. We enriched our dataset by incorporating this feature, effectively increasing the volume of data at our disposal. eGFR serves as a critical indicator for diagnosing CKD and assessing its progression, thereby guiding clinical management. Our focus is on developing machine learning and deep learning models for the efficient and precise prediction of CKD. To ensure the reliability of our approach, we employed robust data collection and preprocessing techniques, resulting in refined information for model training. Our methodology integrates various machine learning and deep learning models, including four machine learning algorithms: adaptive boosting (AdaBoost), random forest (RF), Bagging, and artificial neural network (ANN), as well as a hybrid model. Our proposed ANN_AdaBoost model not only introduces a novel perspective by addressing an identified gap but significantly enhances CKD prediction.