Federated Machine Learning for Privacy-Preserving Cyber Threat Intelligence in Global Cloud-Integrated MIS Platforms

Md Kazi Tuhin (Yeshiva University, Manhattan, New York, USA)
Md Talha Bin Ansar (Yeshiva University, Manhattan, New York, USA)
Mohammad Somon Sikder (Pacific States University, Los Angeles, California, USA)
Hemayet Uddin Himel (Pacific States University, Los Angeles, California, USA)
Harleen Kaur (Westcliff University, Irvine, California, USA)
Hasan Imam (International American University, Los Angeles, California, USA)
Ahmed Shan-A-Alahi (Washington University of Science and Technology, Virginia, USA)

The Eastasouth Journal of Information System and Computer Science
Vol. 2 No. 02 (2024): The Eastasouth Journal of Information System and Computer Science (ESISCS)

Cyber Intelligence (CI) is an advanced security system that defends networks against cyberattacks by using ML models. In conventional centralized machine learning tools used to detect cyber threats, critical organizational information must be shared, which imposes severe privacy and security issues. To overcome this issue, this paper suggests a privacy-aware cyber threat detection system based on Federated Learning combined with a Convolutional Neural Network (FL-CNN). The proposed framework is tested on CIC-IDS-2017 dataset, where the methods of preprocessing like: missing value imputation, MinMax normalization, one-hot encoding, and attack label remapping are used. Experiment findings show that the suggested FL-CNN model outperforms the centralized CNN model in most metrics, including accuracy (96.8%), precision (96.2%), recall (96.7%), and F1-score (96.5%). Also, a false positive rate of the model stands at 0.025, which is much lower meaning that the model is better at identifying a cyber threat in a distributed cloud system.