Hybrid Deep-Ensemble Learning for Cybersecurity: A Multi-Dataset Framework Achieving High Precision and Minimal False Positives in Attack Detection

Effective cyberattack detection is a major challenge in network security systems. One of the problems that often arises is the high false positive rate, which causes a large number of false alerts that burden the security team and the system as a whole. This research aims to develop a machine learning-based hybrid model that can improve the accuracy of attack detection while significantly reducing the false positive rate. The proposed model combines a deep learning approach with ensemble learning to optimize cyber attack classification capabilities. The research method involves a series of experiments with three standard cybersecurity datasets, namely CICIDS 2017, NSL-KDD, and UNSW-NB15. The model was tested using key evaluation metrics, such as precision, recall, F1-score, and false positive rate. The experimental results show that the developed Hybrid model has a precision of 96.1%, recall of 92.5%, and F1-score of 94.2%, with a lower false positive rate of 5.8%. This model proved superior to other approaches, such as Random Forest, XGBoost, and LSTM, which still showed a higher false positive rate. The advantage of the approach used in this research lies in its ability to recognize more complex attack patterns and increase the reliability of the threat detection system. In addition, the results show that this method has consistent performance on various datasets, so it can be widely applied in cybersecurity systems. Thus, this research contributes to developing a more accurate and efficient cyber attack detection method.