Hybrid resampling technique with HWSO based temporal convolution network for credit card fraud detection

In the wake of recent progresses in electronic trade and communiqué links, credit card use has skyrocketed for both online and in-person purchases. Maximum credit card datasets are very skewed, making it difficult to design efficient fraud detection algorithms that can help mitigate these losses. Traditional approaches are inefficient for credit card fraud detection because their architecture requires a vector to the output vector. As a result, they are unable to billet the ever-changing holders. In instruction to well recognize credit card fraud, the authors of this research suggest a hybrid classifier and data resampling strategy. The hybrid resampling is accomplished by combining the synthetic marginal oversampling procedure (SMOTE) with the edited nearest neighbour (ENN) approach. Temporal convolutional networks (TCN) are combined with a Bidirectional Gated Recurrent Unit (BiGRU) and a Dual Attention network (DATT) to perform the categorization in the suggested model. Second, in order to quickly get the deep semantic features of the credit card data, we employed TCN and BiGRU networks to extract characteristics were then spliced and merged, and a dual attention method was implemented to assign global weight to the most crucial information. In the end, classification was performed using a Softmax classifier. The accuracy of the categorization is further enhanced by the use of the Hybrid White shark Optimisation model (HWSO) model for selecting the model’s weights. Using publicly accessible, credit card dealings, the recommended strategy is proved. The trial results show that the models adjusted using the proposed approach outperformed those using hybrids of competing metaheuristics.