A computational intelligence approach for a better diagnosis of diabetic patients

•A computational intelligence approach has been proposed to analyze diabetes patients effectively.•In this framework the biggest strength is identification of false split points and Gaussian fuzzy membership function.•The number of split points that are obtained are minimized by identifying and eliminating the false split points.•The framework is illustrated with the case of Pima Indian diabetes patient’s data. Knowledge discovery refers to identifying hidden and validpatterns in data and it can be used to build knowledge inference systems. Decision tree is one such successful technique for supervised learning and extracting knowledge or rules. This paper aims at developing a decision tree model to predict the occurrence of diabetes disease. Traditional decision tree algorithms have a problem with crisp boundaries. Much better decision rules can be identified from these clinical data sets with the use of the fuzzy decision boundaries. The key step in the construction of a decision tree is the identification of split points and in this work best split points are identified using the Gini index. Authors propose a method to minimize the calculation of Gini indices by identifying false split points and used the Gaussian fuzzy function because the clinical data sets are not crisp. As the efficiency of the decision tree depends on many factors such as number of nodes and the length of the tree, pruning of decision tree plays a key role. The modified Gini index-Gaussian fuzzy decision tree algorithm is proposed and is tested with Pima Indian Diabetes (PID) clinical data set for accuracy. This algorithm outperforms other decision tree algorithms.