Advancing Chronic Kidney Disease (CKD) Diagnosis: A Comparative Analysis of Machine Learning and Deep Learning Approaches

Chronic kidney disease (CKD) is a non-communicable disease in which our kidneys are stop functioning with time. Accordingly, World Health Organization survey reports, during year 1990 to 2023, life loss caused by CKD were increased up to $\mathbf{9 0 \%}$. The growth rate of CKD patients also increasing rapidly and presently 850 million people across world are suffering from CKD. The kidneys are preliminary requirements for well running of our body and its cleans all waste from the blood. It also controls the red blood cells’ production and regulates the blood pressure in the body by producing hormone. The stage5 is most dangerous stage of CKD. Sometime, very difficult to manage the complication of CKD and treatment of CKD is also very costly. The early detection of CKD is very helpful for minimizing economic burden on patients and as well as, very effective for making appropriate treatment well in advance. In this work, machine learning and deep learning classification models are used for early prediction of CKD. The size of dataset contain 400 CKD patients records with 24 clinical parameters. The classification models such as Logistic Regression (LR), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), and Multi-Layer Perceptron (MLP) and MLP with backpropagation, are used to for prediction of CKD. Accuracy, sensitivity, specificity, and F1 score evaluation for each model were performed using $\mathbf{1 0}$-fold cross-validation. The backpropagation model received highest accuracy score value with 86.5 % among all classification models and logistic regression has received lowest accuracy with 74.2%.