Minimising Security Deviations in Software-Defined Networks Using Deep Learning

This study aims to enhance the security of Software-Defined Networks (SDN) byemploying deep learning techniques to detect cyber threats and mitigate attacks. A comprehensive data analysis was conducted, beginning with feature identification and dimensionality reduction using the Gain Information method to filter out redundant features, thereby improving model performance. Additionally, Min-Max normalization was applied to standardize feature ranges, and the SMOTE technique was utilized to balance the dataset and reduce the impact of underrepresented classes. The research compares the performance of three primary deep learning modelsCNN, LSTM, and ANNwith a newly proposed method designed to better differentiate between similar attack categories. The results demonstrate that deep learning models can effectively uncover hidden patterns in network traffic and accurately classify security threats, with the LSTM model particularly excelling in capturing temporal dependencies. While CNN and ANN models showed high accuracy in certain scenarios, they struggled to identify classes with fewer samples, necessitating the use of additional balancing techniques. Conversely, the proposed method showed promise in achieving a balance between accuracy and efficiency, suggesting that further refinement in feature engineering and advanced balancing strategies could enhance its performance. Overall, this study underscores the critical role of integrating deep learning with advanced preprocessing techniques in developing more reliable and effective intrusion detection systems for SDN environments.