Identification of Suspicious Patterns and Prevention of Telephone Fraud Incidents Using a Neural Network

This article describes software for detecting suspicious patterns and preventing telephone fraud incidents using a neural network, and also determines the accuracy of the model for solving the problem of social engineering on the example of the financial sector. The analysis was carried out on the basis of a data set that included 600 copies of the text: 300 conversations of scammers and 300 ordinary conversations on the topic of finance. To ensure the confidentiality of personal data, the text was filtered using a dictionary. The paper used a sequential model of a convolutional neural network with an optimizer “Adam” and a loss function “Categorical_crossentropy”. As a result of product development, the positive and negative aspects of using the proposed solution were identified. The results obtained can be applied to reduce the total number of cases of telephone fraud, as well as increase confidence in banks and the service sector. In the future, the developed software will solve more global problems.