Enhancing the performance of decision tree using NSUM technique for diabetes patients

Abstract

Diabetes is a common disease among children to adult in this era. To prevent the diseases is very important because it saves the human lives. Data mining technique helps to solve the problem of predicting diabetes. It has steps of processes to predict the illness. Feature selection is an important phase in data mining process. In feature selection when dimension of the data increases, the quantity of data required to deliver a dependable analysis raises exponentially. Numerous different feature selection and feature extraction techniques are present, and they are widely used filter-based feature selection method is proposed which takes advantage of the wrapper, Embedded, hybrid methods by evaluating with a lower cost and improves the performance of a classification algorithm like a decision tree, support vector machine, logistic regression and so on. To predict whether the patient has diabetes or not, we introduce a novel filter method ranking technique called Novel Symmetrical Uncertainty Measure (NSUM). NSUM technique experimentally shows that compared to the other algorithms in filter method, wrapper method, embedded method and hybrid method it proves more efficient in terms of Performance, Accuracy, Less computational complexity. The existing technique of symmetric uncertainty measure shows less computational power and high performance, but it lacks in accuracy. The aim of the NSUM method is to overcome the drawback of the filter method, i.e., less accuracy compared to other methods. NSUM technique results show high performance, improved accuracy, and less computational complexity. NSUM method runs in 0.03 s with 89.12% as accuracy by using Weka tool.