ACO-KNN Predictive Model for Diagnosis of Chronic Kidney Disease

Chronic Kidney Disease (CKD) remains a worldwide health challenge that is increasing steadily. It is a chronic situation accompanied by an increase in morbidity, mortality, and also a risk of other several diseases like cardiovascular diseases and high healthcare costs. More than two million individuals over the globe receive dialysis or transplanting kidney treatment to stay alive, yet this figure shows only 10% represent people who need treatment to live. Early detection and management of CKD are necessary. It is important to predict the progression of CKD with reasonable accuracy due to its dynamic and covert nature in the early stages and patient heterogeneity. This paper presents a CKD predictive model by the introduction of a nature-inspired computation algorithm known as Ant Colony Optimization for the selection of discriminant attributes from the CKD indigenous dataset and employing some selected machine learning algorithms for classification. The CKD predicted model was evaluated using an indigenous dataset collected from Ladoke Akintola University of Technology (LAUTECH) teaching hospital, Ogbomoso and Osogbo, University College Hospital (UCH), Ibadan, Oyo State and Obafemi Awolowo University Teaching Hospital (OAUTH), Ile-Ife, Osun State, Nigeria. Experimental results showed that binary classification for CKD predictive model produced the best accuracy of 99.13%, the best specificity of 0.9839, the best sensitivity of 0.9929 in ACO-KNN and also for the multistage CKD predictive model, the best outputs for accuracy, specificity, sensitivity are given respectively with 99.65%, 0.9956 and 1.000 in CKD patients with stage 2 disease Severity using ACO-KNN.