Intrusion detection system (IDS) using machine learning

Tan, May May (2024) Intrusion detection system (IDS) using machine learning. Final Year Project, UTAR.

Preview

PDF Download (4Mb) | Preview

Abstract

Intrusion Detection Systems (IDS) play a critical role in safeguarding organizational networks by identifying potential threats and anomalies. However, traditional IDS approaches often suffer from high false alarm rates, leading to unnecessary alerts for administrators. This research focuses on enhancing machine learning and deep learning-based IDS models to improve accuracy, precision, recall, and F1 score, ultimately aiming for a more balanced and effective performance. The study leverages the CIC-IDS2017 dataset for model training, employing Random Forest (RF), Deep Neural Network (DNN), and Deep Autoencoder (DAE) architectures. A rigorous pre-processing phase, including data cleaning and feature selection using Pearson’s Correlation, enhances the dataset's quality and relevance. Subsequently, the refined dataset undergoes model training and testing. Hyperparameter tuning, facilitated by grid search, fine-tunes key features to optimize model performance. Evaluation metrics such as accuracy, precision, recall, and F1 score are employed to assess the models' efficacy in binary and multi-class classification tasks. Results demonstrate significant improvements and balanced performance compared to previous research models, achieving an average performance of 99.5% across all models.

Actions (login required)