ПРИМЕНЕНИЕ НЕЙРОСЕТЕВЫХ МОДЕЛЕЙ ДЛЯ АНАЛИЗА КЛАВИАТУРНОГО ПОЧЕРКА В ИНФОРМАЦИОННОЙ БЕЗОПАСНОСТИ

Abstract

This paper presents a systematic review of neural network models used for keystroke dynamics analysis in biometric user authentication. The architectures of LSTM, GRU, CNN, transformers, and Siamese networks are examined, along with an analysis of their strengths and limitations. Special attention is given to hybrid models, which demonstrate improved accuracy and robustness to changes in user behavior. The results of a comparative analysis based on public and user-specific datasets are provided, and key performance metrics such as accuracy, FAR, and FRR are discussed. It is shown that neural network methods significantly enhance authentication reliability, but further research is needed in the areas of adaptability and biometric data protection. The study highlights the potential of neural network solutions in improving the security level of modern information systems.