Date : 23/06/2025
Hour: 20:15
Student Name & ID: LLAHM OMAR FARAJ OMAR // STUDENT ID : (245105402)
Supervisor: Prof. Dr. Ömer KARAL
Link or Room: https://meet.google.com/hhb-bmhu-ggb
Meeting Link (Online)
OR
Room Number (Face to Face)
Abstract:
The escalating complexity and volume of modern network traffic present formidable challenges for effective monitoring, anomaly detection, and security enforcement. Traditional analytical methods often falter in discerning subtle deviations indicative of malicious activities or performance bottlenecks. This research introduces a novel approach leveraging deep Long Short-Term Memory (LSTM) networks within a multitask learning framework for comprehensive network traffic analysis and anomaly detection. By treating network flow data as intricate time series, the proposed system concurrently addresses both classification (e.g., identifying anomalous patterns) and regression (e.g., predicting future traffic metrics) tasks. Utilizing the CESNET-TimeSeries24 dataset, the study meticulously evaluates the model's efficacy across various performance indicators, including a detailed analysis of classification metrics, regression accuracy, and the visualization of key network parameters over time. The findings underscore the profound capability of deep LSTMs in extracting temporal dependencies and intricate patterns from high-dimensional network data, thereby offering a robust and adaptable solution for enhancing network security and operational intelligence.
