An Effective and Accurate Fusion Result from Multi Class Ensemble Classification

Financial fraud detection is a most challenging task in an online transaction oriented applications which concern more to provide the secured environment for the users. Various researches has been conducted previously that focus on providing the most secured environment to the users by finding and preventing the malicious patterns. Classification is one of the most proved techniques for detecting the most malicious patterns that resides in the financial database by using which the malicious patterns can be identified. In our previous research work Optimal Ensemble Architecture Selection using Firefly and the dempster shafer theory based Ensembling is done for finding the fraudulent behaviour in the accurate manner. The ensemble classifier fusion approach used in the previous methodology called dempster shafer theory retrieves the fusion result as classifier output with more confidence value. This approach is computationally inefficient and doesn’t concentrate on interrelation between different classifier results due to its additive measure property. This problem is resolved in this work by introducing the fuzzy integral measure based ensemble fusion using sugeno integral (FIM-EFSSI) and the fuzzy integral measure based ensemble fusion using Choquet integral (FIM-EFSCI). These approaches can find the better and accurate Ensembling result by considering the relation between the different classifier results. The experimental tests conducted were proves that the proposed approach provides better result than the existing approach in terms of improved classification accuracy in the matlab simulation environment.