Hybrid Optimization and Deep Learning for Detecting Fraud Transactions in the Bank

Abstract

This paper, the efficient and effective fraud detection technique, termed SpiHWO-based Deep RNNtechnique is developed. At first, the data transformation is performed for transforming the data using Yeo-Johnson transformation. After that, the effective features are selected based on wrapper method where the best features are selected for further processing. Then, by using selected features, fraud detection is executed based on Deep RNN classifier, which is trained by developed SpiHWO technique. The proposed SpiHWO algorithm is newly developed by combining the SMO algorithm and HWO algorithm. Furthermore, the performance of developed method is computed using performance metrics, like sensitivity, specificity and also accuracy. The developed method achieved improved performance with respect to accuracy of 0.951, sensitivity of 0.985 and specificity of 0.792.