The characteristics of frauds are unstable, and they require unsupervised learning since they are always changing the way they behave. Scammers can conduct online transactions by using the most recent technologies available to them. Fraudsters assume the habits of their victims, and their schemes rapidly gain traction. Although some criminals use online channels first before moving on to other techniques, fraud detection systems must apply unsupervised learning to identify online payments. Particularly, focus on fraud scenarios that cannot be detected with previous records or supervised learning. This work aims to create a support vector machine model (SVM) to detect anomalies from regular pattern produced by competitive swarm optimisation (CSO). Therefore, CSO-SVM method is implemented based on unsupervised learning utilizes for effectively detect credit card fraud. Implemented CSO-SVM method obtains high performance when compared to existing methods including Competitive swarm optimization and Deep convolutional neural network (CSO-DCNN), dilated CNN (DCNN), and coarse K-neural network (KNN), and result achieved value of 99.88% accuracy.