Comparative performance of urine and bloodbased attributes in chronic kidney disease prediction using artificial neural network

Chronic Kidney Disease is one of the leading causes of death worldwide. An intelligent diagnostic system capable of early detection is therefore, becoming increasingly important. These would allow effective intervention to be delivered to patients, thus prolonging kidney function and reducing risk of mortality. The system should be non-invasive, convenient, accurate, and reliable in detecting the required attributes. This study compares between urine- and blood-based attributes in acute renal failure prediction using artificial neural network. A total of 400 sample data is obtained from UCI Machine Learning Repository. Multiple imputation is then implemented to generate synthetic data. These overcome the issue of missing datapoints and unbalanced sample distribution. Two artificial neural network models are trained. One using the urine-based attributes and the other, using blood-based attributes. Both models attained excellent classification accuracies of 96.0% and 98.0%, respectively. However, the ANN model developed based on urine-based attributes are recommended intelligent diagnostic systems due to lower computational requirements and the sample acquisition protocol is much convenient for patients and medical practitioners