Abstract
There has been an increase in credit card fraud as e-commerce has become more widespread. Financial transactions are essential to our economy, so detecting bank fraud is essential. Experiments on automated and real-time fraud detection are needed here. There are numerous machine learning techniques for identifying credit card fraud, and the most prevalent are support vector machine (SVM), logic regression, and random forest. When models penalise all errors equally during training, the quality of these detection approaches becomes crucial. This paper uses an innovative sensing method to judge the classification algorithm by considering the misclassification cost and at the same time by employing SVM hyperparameter optimization using grid search cross-validation and separating the hyperplane using the theory of reproducing kernels like linear, Gaussian, and polynomial, and the robustness is maintained. Because of this, credit card fraud has been identified significantly more successful than in the past.