Research Authors & Affiliations

Abdullah Rakib Akand¹, Md. Mehedi Hasan Bhuiyan Nipu², Isha Das³, Golam Ali⁴, Atahar Hossain⁵, & Fahad Siddique Faisal⁶

¹Dept. of Computer Science and Engineering, Asian University of Bangladesh, Dhaka, Bangladesh
²Dept. of Computer Science and Engineering, North South University, Dhaka, Bangladesh
³Network Communication and IoT Lab, Chittagong University of Engineering and Technology, Chattogram, Bangladesh
^4,5Dept. of Computer Science and Engineering, University of Science and Technology Chittagong, Chattogram, Bangladesh
⁶Electronics and Telecommunications Engineering, Chittagong University of Engineering and Technology, Chattogram, Bangladesh

Conference Information

Published in: 2026 5th ICECTE
Publisher: IEEE | Location: Rajshahi, Bangladesh
Conference Date: 29-31 January 2026
Added to IEEE Xplore: 16 March 2026
DOI: 10.1109/ICECTE69292.2026.11429379

Addressing the security of an organization from internal zero-day threats is highly problematic for classical security approaches, as such vulnerabilities are unknown and hence evasive are conventional detection methods. For zero-day threats, Machine Learning (ML) is a game-changer for enhancing the detection capabilities of intrusion detection systems (IDS). Nevertheless, many existing ML-based IDS architectures are still confronted with class imbalance, feature selection, overfitting, latency, and other issues that diminish real-time use. My goal is the creation of a ML based IDS that is robust and portable to flexible environments, as well as automated to determine under latency constraints zero-day attack detection. Dealing with recent network traffic datasets, a comprehensive preprocessing framework to address missing values, feature selection, and balance the dataset with SMOTE is constructed. Predictive models such as Logistic Regression, Random Forest, XGBoost, LtGBM, Neural Networks (NN), and Convolutional Neural Networks (CNN) are developed, and performance is examined through multiple metrics-accuracy, precision, recall, F1 and balanced accuracy. LGBM and deep learning models demonstrate the best performance, which justifies the need for advanced zero-day intrusion prevention.