Identification of Anomalies via Deep Learning-Based Models for High-Dimensional Telecom Traffic Data

The rapid growth in size and complexity of today’s telecommunication networks has created opportunities for networks to be attacked via anomalous behaviors, adversely impacting the security, reliability and performance of network services. The traditional approaches for identifying anomalies in networks (Rule-Based Anomaly Detection) do not perform as well when applied to high-dimensional, dynamically changing traffic patterns, thus limiting their applicability to real-world environments. Suggested a novel Deep Learning-based architecture for identifying traffic irregularities in telecommunications networks in order to address this issue. The framework includes a comprehensive set of data preprocessing techniques, such as Feature normalization of the data values and techniques for balancing the classes within the Data Set, that allow for the creation of unique models capable of discriminating between normal and anomalous traffic. developed and evaluated two different deep learning architectures for the problem; an Artificial Neural Network (ANN) and a 1D Convolutional Neural Networks (CNN) using the Network Traffic Anomaly Detection Dataset (Kaggle). The experimental results show that both of the models perform significantly better than traditional machine learning techniques, i.e., ANN achieved 95.27% accuracy as well as best F1-Score while the CNN detected the temporal traffic patterns. The recall values were consistently high across both architectures, indicating that both architectures are capable of detecting anomalous events reliably. Therefore, the proposed framework shows promise for the development of scalable Real-Time Telecommunication Network Anomaly Detection Systems.