BLOCKCHAIN FRAUD TRANSACTION FOR FRAUD DETECTION IN BANKING DATA

Abstract

The use of credit cards for purchases and other financial activities grew in
tandem with the proliferation of e-commerce services and other technological advancements.
High bank transaction fees are necessary due to the evident rise in fraud. Therefore, identifying
fraudulent activity has become an intriguing subject. To regulate the relative weights of fake and
legitimate transactions, this study investigates how category weight-tuning hyperparameters
might be used. In order to solve real-world problems like imbalanced data and optimize the
hyper parameter values, we use Bayesian optimization. If we want to make the Light GBM
method work better by taking the voting mechanism into consideration, we need pre-process
imbalanced data using X G Boost and Cat Booster on top of weight-tuning. They use deep
learning to refine the hyperparameters, namely the one we suggest—weight-tuning—to further
optimize performance. To ensure the suggested approaches work, we conduct several trials using
actual data. In addition to the classic ROC-AUC, recall-precision measurements are used to
better cover imbalanced datasets. Different versions of XG Boost, Light GBM, and Cat Boost are
tested independently via a 5-fold cross-validation procedure. Using an overwhelming ballot
ensemble learning approach, we may evaluate the coupled algorithms' performance even
further. Light GBM & XG Boost meet the optimal level requirements with ROC-AUC = 0.95,
preciseness 0.79, recall 0.80, F1 rated 0.79, and MCC 0.79– according to the findings.
Employing deep neural networks in conjunction with the Bayesian optimization method also
yields the following results: With an F1 rating of 0.81, an MCC of 0.81, an accuracy of 0.80, and
a recall of 0.82, the ROC-AUC is 0.94. Compare this to the status quo, and you'll see a huge
improvement.