Improving the multilayer Perceptron neural network using teaching-learning optimization algorithm in detecting credit card fraud

Due to the necessity of electronic transactions with credit cards in this modern era and that fraudulent activity with credit cards are on the rise, the development of automated systems that can prevent such financial fraud is considered vital. This study presents a method for detecting credit card fraud by deploying a neural network that distinguishes between legitimate and illegitimate transactions and detects fraudulent activities with stolen physical credit cards. For this purpose, after collecting data in the preprocessing stage, cleaning and normalizing the data, the feature selection operation is performed using fisher discriminant analysis. After that, a multilayer perceptron (MLP) neural network is trained during the post-processing period using the teaching learning-based optimization algorithm (TLBO) to optimize credit card fraud detection. In this algorithm, local search (exploitation) is done using the teacher phase, and global searching(exploration) is done using the student phase. Moreover, the fisher discriminant analysis algorithm reduces within-class scattering. It increases between-class diffusion to increase classification accuracy and decrease the CPU time of the algorithm in the training phase. The latest available algorithms such as AdaBoost, Random Forest, CNN, and RNN are also compared with the proposed method. The results show that the proposed algorithm outperforms the mentioned algorithms regarding some standards criteria and CPU time.