Resumen
Today there is a significant increase in the number of incidents involving the use of plastic cards, as well as a variety of fraudulent methods used by cybercriminals. The presented article is devoted to the application of machine learning methods to counter fraudulent transactions using plastic cards. The aim of the article is to study the effectiveness of various machine learning models in the analysis of transactions with plastic cards to identify various types of fraud. The article sequentially analyzes such machine learning methods as RandomForest, CatBoost, LogisticRegression, 2-layer ordinary perceptron and Rumelhart's multilayer perceptrons L-BFGS and SGD. Considerable attention is paid to the process of preparing data for participation in modeling, which is iterative and includes operations for selecting tables, attributes, records, converting, cleaning data, filtering and combining them in the desired format. The dataset was taken from Worldline and the ULB Machine Learning Group for Big Data Mining and Fraud Detection. The problem with class imbalance was solved using resampling. Rumelhart's L-BFGS and SGD perceptrons showed the best results in experiments when using the metrics "response time" and "accuracy".