Quantitative Study of Neural Network-Based Anomaly Detection for Real-Time Cyberattack Identification in Large-Scale Computing Networks

The objective of this quantitative study was to evaluate the effectiveness of neural network–based anomaly detection models—particularly temporal architectures—in accurately identifying cyberattacks in real-time within large-scale computing networks. A comprehensive network traffic dataset containing both normal and malicious flows was preprocessed and used to train multiple deep learning architectures, including feedforward neural networks, convolutional neural networks (CNN), recurrent neural networks (RNN), and long short-term memory (LSTM) models. Feature normalization, sequence transformation, and train–test splitting were applied before training. Model performance was assessed using accuracy, precision, recall, F1-score, false positive rate, and detection latency. Traditional machine learning models such as logistic regression, support vector machines (SVM), and random forest were also evaluated for comparison. The LSTM-based anomaly detection model demonstrated the highest performance, achieving a validation accuracy of 97.5%, precision of 96.8%, recall of 95.9%, and an F1-score of 96.3%. The average detection latency was 18.4 ms, confirming real-time capability. While other neural network models performed competitively, traditional machine learning methods showed significantly lower accuracy and slower processing speeds. The LSTM model also achieved a low false positive rate of 2.1% and demonstrated strong generalization when tested on unseen traffic conditions. Neural network–based anomaly detection, particularly using LSTM architectures, provides a highly effective and scalable solution for real-time cyberattack identification in large-scale networks. The study highlights the superiority of deep learning over traditional machine learning methods in capturing complex temporal and nonlinear patterns in network traffic. Future research should investigate adversarial resilience and computational optimization to further enhance deployment in operational cybersecurity environments.