HOMOMORPHIC ENCRYPTION-ENABLED DEEP LEARNING MODEL FOR INTELLIGENT CYBER THREAT DETECTION

Cybersecurity is one of the emerging fields which address information security,protecting computer systems and networks from intrusions.Cyber-attack detection is crucial for safeguarding computer networks against unauthorized access and malicious activities. Traditional machine learning approaches often suffer from overfitting, leading to reduced accuracy.To overcome this, an optimized deep learning approach with a homomorphic encryption based authentication protocol is proposed. Cyber threat data is collected from open sources and pre-processed using data cleaning and binning method. Feature extraction is carried out using TF-IDF, followed by dimensionality reduction through Principal Component Analysis. The processed data is then classified using an Extended Physics-Informed Neural Network (EPINN) for cyber-attack detection. A Signature-based Authentication Protocol ensures secure user authentication, while BFV encryption secures data storage in the cloud. Experimental results show high efficiency with 97.36% accuracy, 89.78% precision, and a low false positive rate of 2.17%. This approach enables automatic and robust cyber threat detection, improving proactive defense mechanisms for organizations.