Blockchain-Based Private AI Model with RPOA Based Sampling Method for Credit Card Fraud Detection
S. Stephe, Valureddi Revathi, B. Gunapriya, Arunadevi Thirumalraj · 2024
Credit cards are among the most popular ways to pay for things online in many countries, both developed and developing. Online purchases have been simplified, made easier, and improved with the development of the credit card. The rate of fraud has increased, though, since it has provided thieves with additional opportunity to conduct fraud. Many organisations and enterprises depend significantly on machine learning techniques to notice and automatically categorise fraudulent connections due to the high volume of transactions. There is a serious problem with the data imbalance since machine learning technique effectiveness is highly dependent on the training data quality. As a whole, the data only shows a tiny fraction of fraudulent transactions. As a result, machine learning classifiers are significantly less effective. More than that, the most secure way to combine banking is being promoted by using blockchain technology. There is an annual uptick in fraud, nevertheless, coinciding with the rise of these sophisticated tools. In light of the low frequency of actual fraudulent transactions, this research presents a novel data augmentation methodology to rectify the data imbalance that has long plagued efforts to identify credit card fraud. By mimicking the actions of real red pandas, the Red Panda optimisation algorithm (RPOA) finds the sweet spot for tuning the hyperparameter of the suggested model. The suggested RPO method is mathematically represented in two stages: exploration, which is based on red panda foraging strategy simulation, and exploitation, which is based on red panda climbing tree movement simulation. The suggested approach’s key benefit is that it eliminates the requirement for parameter modification by mathematically describing the problem without a control parameter. Finally, deep learning algorithms are coupled with blockchain technology to detect fraudulent transactions in the network. The suggested methods are subsequently tested against a number of the most popular classification algorithms. When compared to alternative categorisation approaches, these findings demonstrate that the projected model has the best Precision and Recall.