Development of an Intrusion Detection Model using Long Short Term Memory Algorithm

Abstract

Intrusion detection model is a cyber-security technology that is made to keep an eye on and examine network or system activity for indications of unauthorized or hostile activity. It functions by looking through system logs or network traffic and comparing the data with predefined signatures or patterns of activity that point to possible security risks. This study is aimed at developing deep learning model specifically tailored for Intrusion Detection Systems (IDS) by exploring parameter tuned LSTM (Long-short-Term Memory) network that analyzes datasets by carrying out sequential learning and help to improve traditional IDS to identify threats more accurately and classify intrusions into DoS (Denial of Service), Probe, R2L (Remote to Local), U2R (User to Root), focusing on minimizing false positives and false negatives through continuous model refinement; implement mechanisms for continuous learning to stay ahead of emerging threats using appropriate metrics to access the performance of enhanced IDS. The implementation was carried out on Google Colab platform with 8GB RAM hard ware using python programming language. The results of training and validation accuracy and loss respectively of 10 epochs are recorded. Optimal training and validation accuracy are obtained at 9th and 10th with 99.45% and 99.65% respectively. The results also showed, training and validation loss of 0.0730 and 0.020 respectively. At these optimal validated results, corresponding confusion matrix are obtained and sorted and, accuracy, specificity, f1 score, precision and recall of 99.65%, 99.35%, 99.59%, 99.29% and 99.9% respectively are obtained and compared with other existing models. The initiative of this work aimed at reducing false positives and false negatives, give security analysts more dependable alerts, and make resource allocation more effective.