A NEURAL NETWORK BASED DATA MINING APPROACH FOR RECOGNITION OF CHRONIC KIDNEY DISEASE

Chronic kidney failure occur when the regular kidney filtration functionalities fails which leads to accumulation of electrolytes, wastes and other fluids in the body. One has to go appropriate dialysis procedure for their survival. It is very critical to recognize the level of chronic kidney disease (CKD) for the nephrologist and further dialysis period cannot be predicted appropriately for individuals. Data mining approaches have shown a promising path in the last decade to develop automated decision making tool for clinical diagnosis. This specific research suggests the application of neural network as critical qualitative indicator to mine the kidney dialysis attributes for classification of CKD from non-chronic kidney disease (NCKD). Two datasets one from open source UCI machine learning repository CKD database and other local hospital were considered for this study. Initially clustering was applied to remove the inconsistency from the datasets. Numerical and nominal normalized data was employed to multilayer perceptron neural network (MLPNN) to perform the classification of CKD and NCKD.MLPNN was configured optimally by appropriate network parameters and was evaluated in terms of Specificity, Sensitivity and classification accuracy. Further other classifier performance metrics, such as, position and negative predictions, error rate, F-Score, MCC and Kappa test were also evaluated. Experimental simulation shows that the proposed pattern classifier yields a classification accuracy of 93.22% and 92.78% respectively for the two different data sets.