Enhanced APT Attack Detection Using Convolutional Neural Networks and Deep Learning Models

One type of cyberattack known as Advanced Persistent Threats (APTs) occurs when an unauthorized user enters a computer system without authorization and remains hidden for a long period of time. Instead of destroying the network system, they targeted extremely valuable and sensitive data to steal it. These threats are dangerous for banks, government security, and business information. This study is conducted to evaluate and comparative analysis of deep learning models to identify the best one for detecting APT attacks. Each model has different strengths, and combining them helps improve accuracy and detection. The models used for testing are LSTM, CNN,AE, and hybrid models, in which LSTMs excel at capturing long-term dependencies in data, CNNs are highly effective at analyzing spatial features, and AEs help detect anomalies by reconstructing input data. By combining these models, the system gains the strengths of all three, enhancing its ability to detect complex attack patterns. from which the AE + CNN + LSTM (hybrid) model performed best among all other models in all four metrics - Precision, Accuracy, F1-Score, and Recall. In conclude, AE + CNN + LSTM (hybrid) is the most suitable model for APT attack detection.