Detection of anomalies in networks by implementing and analyzing deep learning-based classifiers

With the increasing complexity and sophistication of cyber threats, the demand for effective anomaly detection mechanisms in network security has grown substantially. This research focuses on the utilization of deep learning (DL) to detect anomalies in networks, implementing and analyzing three distinct models of deep learning-based classifiers. The classifiers include a binary classification model, a multiclass classification model with proportionality classification, and a multiclass classification model with exclusion strategies. The efficacy of these models is evaluated through cross-validation runs and testing on data from a database parallel to that of its training. The binary classification model serves as the baseline, distinguishing between normal and anomalous network behavior. The multiclass classification model with proportionality classification extends the binary approach by categorizing anomalies into multiple classes with consideration for their proportional representation. On the other hand, the multiclass classification model with exclusion strategies refines anomaly detection by explicitly excluding certain patterns from classification, enhancing the model's specificity. The research methodology involves the development of deep learning-based classifiers, the collection of network data for training, and rigorous evaluation through cross-validation techniques. Additionally, the models are tested on a separate dataset to assess their generalization capabilities in real-world scenarios. The outcomes of this research contribute to the understanding of DL applications in network anomaly detection, providing insights into the strengths and limitations of different classifier strategies.