A Novel Algorithm to Diagnosis Type II Diabetes Mellitus Based on Association Rule Mining Using MPSO-LSSVM with Outlier Detection Method

Abstract

Background/Objectives: The execution of Frequent Pattern Growth algorithm on medical data is difficult. Association rule based classification is an interesting area focused that can be utilized for early diagnosis. Methods/Statistical analysis: Discretization phase is necessary to transform numerical characteristics. The results are given to Complete Frequent Patten Growth++ for the purpose of rule induction. Accordingly, using Modified Particle Swarm Optimization together with Least Squares Support Vector Machine scheme (MPSO-LSSVM) rules are produced with outlier detection method. Pima Indians Diabetes Data Set is taken as an input. The execution time, number of rules generation and the detection of outlier percentage are analyzed. Results: The CFP-growth algorithm utilizes for finding frequent patterns where constructing the Minimum Item Support (MIS)-tree, CFP-array and producing frequent patterns from the MIS-tree. From the set of frequent item sets found, create all the association rules that have a confidence exceeding the minimum confidence. The Enhanced outlier detection method is used for determining the outlier degree from association rules for outlier detection. Association rules are mined using MPSO-LSSVM classification based association rule mining algorithm. The classification based association rule generation using MPSO-LSSVM is utilized first time in this work with outlier detection method. For the reason of eradicating the effect of unavoidable outliers in investigation sample on a scheme’s performance, a new MPSO-LSSVM with the integration of outlier detection method is proposed first time. The experimental observations reveal that this framework provides a better accuracy of 95% when evaluated against the existing techniques. Conclusion/Application: CFP-Growth++ proposed for rule pruning and MPSO-LSSVM based algorithm used for mining association rules from Type-2 DM dataset. This work is suitable for early detection of type-2 diabetes mellitus disease.