Detection of Advanced Persistent Threat Attacks With Optimized Deep Learning Framework
Indra Kumari, Hansung Lee · IEEE Access · 2025
Advanced Persistent Threat is considered as one of the most harmful security attacks that becomes a challenge for the governments, business sectors, and organizations. Due to its dynamic nature, the organizations are facing difficulties in determining the exact type of attack. To address this problem, this research paper presents a novel, optimized hybrid deep learning model. The projected model includes four major phases: “(a) pre-processing, (b) feature extraction, (c) feature selection, and (d) attack detection”. Initially, raw data is collected and pre-processed using a data cleaning algorithm. Statistical characteristics, including proposed weighted entropy and higher-order statistical features such as skewness and kurtosis, are retrieved from the pre-processed data during the feature extraction step. The most optimal features are selected from the extracted characteristics utilizing the newly proposed Aquila Customized Sunflower Optimization Algorithm (ACSOA). The Aquila Optimizer (AO) and the standard sunflower optimization (SnFO) algorithms are combined in this ACSOA model. Finally, the attack detection and classification are carried out in the newly built optimized hybrid classifier framework, which is made up of optimized Recurrent Neural Network (RNN) and Gated Recurrent Network (GRU). The weight of the RNN will be fine-tuned using ACSOA to improve the detection accuracy of the projected model. The type of APT assault in network traffic will be depicted in the final detection result using the improved hybrid classifier architecture. Finally, the efficiency of the anticipated model is validated by a comparative examination.