CNN based transfer learning with hybrid optimization for advanced persistent threat detection

In recent years, the Advanced Persistent Threat (APT) has been considered as an emerging sophisticated cyber-attack and is becoming a new security risk to governments and companies. In general, APT is performed by attackers to get access to confidential information. Various methods are employed to detect APT attacks by analyzing their anomalous and indicative behaviors. Meanwhile, these methods encountered many issues due to a lack of typical data from attack vectors. Thus, a Northern Smart Flower Cosine Algorithm_Deep Convolutional Neural Network + Transfer Learning (NSFCA_DCNN + TL) model is introduced for detecting APTs. The input log data is allowed for data normalization using quantile normalization, and the fusion of features from normalized data is executed using fuzzy with different distance measures. Following this, the fused features are augmented using the oversampling method. Then, the APT is detected using DCNN with TL (DCNN+TL) by initializing hyperparameters using a Parallel Convolutional Neural Network (PCNN). Here, the parameters are fine-tuned effectively by using the Northern Smart Flower Cosine Algorithm (NSFCA). Additionally, the NSFCA_DCNN + TL attained effectual outcomes with Negative Predictive Value (NPV), True Negative Rate (TNR), False Positive Rate (FPR), True Positive Rate (TPR), precision, and accuracy of 95.58%, 96.17%, 3.82%, 96.32%, 96.54%, and 95.86%.