Hybrid Deep Learning Enhanced with A Novel Feature Extraction Technique for Detection and Mitigation of Cyber Threats

• Author(s): Asogwa T. C. ; Ezeh Ebere M.
• Paper ID: 1709699
• Page: 841-850
• Published Date: 22-07-2025
• Published In: Iconic Research And Engineering Journals
• Publisher: IRE Journals
• e-ISSN: 2456-8880
• Volume/Issue: Volume 9 Issue 1 July-2025

Abstract

This paper proposes a new real-time feature detection and mitigation framework of cyber threats that will be based on hybrid deep learning technique. The system proposed uses Convolutional Neural Networks (CNN) as the part that extracts the spatial features and Autoencoders (AE) as the part that reduces the number of dimensions and detect anomalies resulting in a CNN enhanced with AE architecture that is optimized and can be used in cyber security applications. One of the main novelties of the system is the usage of a feature extraction method based on cross-correlation and it manages to dynamically select the most appropriate network traffic features by assessing the inter-feature relationships over time which provides the model with flexibility in response to a changing trend of threat and discards duplicating or noise data. The system used Python with TensorFlow and Keras to implement deep learning and tested on a virtualized platform with both synthetic. Evaluation terms showed good results, the training accuracy was 89.23%, validation accuracy was 86.74%, and minimal classification loss. The findings have shown that the CNN and AE in combination with feature extraction of the cross-correlation method has great preference to the accuracy and efficiency of network threat detection systems. The framework provides flexible and scalable real-time cybersecurity defence that can be used to reduce false positives and at the same time guarantee prompt mitigating effects.

Keywords

Cybersecurity; Convolutional Neural Network (CNN); Autoencoder (AE); Cross-Correlation Feature Extraction; Real-Time Threat Mitigation

Citations

IRE Journals:
Asogwa T. C. , Ezeh Ebere M. "Hybrid Deep Learning Enhanced with A Novel Feature Extraction Technique for Detection and Mitigation of Cyber Threats" Iconic Research And Engineering Journals Volume 9 Issue 1 2025 Page 841-850

IEEE:
Asogwa T. C. , Ezeh Ebere M. "Hybrid Deep Learning Enhanced with A Novel Feature Extraction Technique for Detection and Mitigation of Cyber Threats" Iconic Research And Engineering Journals, 9(1)