INTRUSION DETECTION SYSTEM BASED ON MACHINE LEARNING

ABSTRACT The escalating complexity of cyber threats has intensified the need for intelligent, adaptive Intrusion Detection Systems (IDS) capable of identifying both known and novel attacks in real time. Conventional IDS approaches—particularly signature-based and anomaly-based models—have demonstrated significant limitations, including an inability to detect zero-day exploits, elevated false positive rates, and insufficient scalability when confronted with high-volume network traffic. To overcome these deficiencies, this study presents the design and implementation of an IDS leveraging Long Short-Term Memory (LSTM) networks, which are adept at modeling temporal dependencies within sequential network data and enhancing detection precision through deep learning methodologies. The proposed system incorporates a comprehensive preprocessing pipeline that normalizes, encodes, and reshapes raw network traffic into time-series sequences optimized for LSTM input. The LSTM architecture itself comprises multiple hidden layers, dropout regularization mechanisms, and fully connected output layers, collectively engineered to maximize classification performance while mitigating overfitting risks. Empirical evaluation yielded promising results, with the model achieving an accuracy of 88.95% on benchmark test datasets and 82.52% under real-time simulation conditions. These findings underscore the system’s efficacy in discerning both benign and malicious network behaviors, significantly reducing false alarms and enabling timely threat detection. To facilitate operational deployment, the system integrates a user-centric Streamlit interface that supports dynamic visualization of performance metrics, including confusion matrices, ROC curves, and classification reports. This interface enhances interpretability and supports informed decision-making by network administrators. Overall, the research demonstrates the viability of LSTM-based IDS frameworks as robust, scalable, and adaptive solutions to the evolving landscape of cybersecurity threats. By harnessing the strengths of temporal learning and deep neural architectures, the system represents a substantive advancement toward next-generation intrusion detection technologies tailored for modern digital ecosystems.