•  
  •  
 

Document Type

Article

Keywords

Advanced Persistent, Threats (APTs), federated learning (FL), Exploratory Data Analysis (EDA), Gated Recurrent Unit (GRU) model, Convolution neural networks (CNN) model

Abstract

Rapid internet expansion and global cloud storage use have heightened the risk of stealthy, persistent, multi-stage Advanced Persistent Threats (APTs). Distributed and resource-limited cloud environments make identifying these stealthy and dynamic threats difficult for traditional Intrusion Detection Systems (IDSs). This paper present FedNN-APT, a Federated Learning (FL) and hybrid Neural Network (NN) APT detection system to overcome these difficulties. High detection accuracy and distributed, privacy-preserving training over several cloud devices are achieved by this approach. FedNN-APT integrates Gated Recurrent Units (GRUs) and Convolutional Neural Networks (CNNs) to learn temporal and spatial APT behavior features effectively. The framework trains local models on partitioned datasets using GRU-CNN, 1D-CNN, and GRU-Recurrent Neural Network (RNN) models, then selects the optimal model for federated aggregation. The final global model is collaboratively built while preserving data confidentiality. The system is evaluated using an APT Malware dataset consisting of 11,107 samples. Experimental findings reveal that the hybrid GRU-CNN model outperforms other models with an average accuracy of 0.9977, precision of 0.9989, recall of 1.00, and F1-score of 0.9988. The federated model has 0.99global accuracy across four clients. A comparative evaluation with current APT detection systems underscores the superiority of FedNN-APT, especially regarding detection accuracy and flexibility in resource-limited environments. Finally, in this paper introduced FedNN-APT, an advanced approach that combines federated learning and neural networks for the detection of APT attacks while enhancing data privacy via a cloud environment and the findings indicate that incorporating deep learning models into a federated learning framework offers a promising direction for future study in safe and scalable threat detection in cloud systems.

Download

Share

COinS