Objective: To provide the secured and flexible environment which can overcome the current shortcomings and at the same time offer better identification of fraudulent behaviour in the optimized manner.
Method: Two optimization approaches namely Particle Swarm Optimization (PSO) and Fire Fly approach (FFA) are used for optimal selection of ensemble classifier architecture. The different base classifiers used in this work are ADTree, Cart, Prism and Ripper. In our previous research work we attempted to use Optimal Ensemble Classification with PSO (OEC-PSO) for improved detection of financial fraudulent activities. OEC-PSO proceeds with the single ensemble architecture to obtain better classification result by changing combination of classifier and the subset feature in every iteration. This is resolved in this research work by introducing the Optimal Ensemble Architecture Selection using PSO (OEAS-PSO) which would construct different ensemble classifier architecture which would be changes randomly in every iteration along with combination of classifier and subset feature. However, PSO lacks from the performance degradation while selecting the better ensemble classifier in case of presence of more noises such as redundancy. It will lead to more number of iteration, thus the computation overhead would be increased. This is resolved by introducing the Optimal Ensemble Architecture Selection using firefly approach (OEAS-FFA). The firefly approach overcomes the issues of PSO by selecting the optimal ensemble architecture that can provide accurate classification result. Finally, the weighted average fusion method is applied on the selected optimal ensemble classifier to retrieve the final result.
Results: The overall research of this work is evaluated in the Matlab simulation environment to find its performance improvement. This evaluation is conducted between the approaches called the OEC-PSO, OEAS-PSO and OEAS-FFA. The performance evaluation is conducted between these approaches in terms of performance measures called the accuracy, precision and recall.
Conclusion: This analysis work is conducted on the UCI data set from which it is concluded that the OEAS-FFA provides optimal result than the other approaches in terms reduced misclassification cost. The findings of this work demonstrate that the proposed research OEAS-FFA provides better result than the previous approaches.
Keywords
