DESIGN FRAMEWORK MODEL FOR NETWORK INTRUSION DETECTION WITH FEATURE SELECTION ALGORITHMS USING MACHINE LEARNING

Abstract

Traditional gadget mastering-based intrusion detection structures frequently depend upon a unmarried set of rules, main to barriers which include inflexibility, low detection rates, and inadequate handling of excessive-dimensional information. To deal with the ones stressful conditions, this paper proposes a singular widespread intrusion detection framework along with 5 components: a preprocessing module, an autoencoder module, a database module, a category module, and a remarks module. The preprocessing module prepares the information, this is compressed through autoencoder module to generate lower-dimensional reconstruction abilities, permitting class module to offer accurate effects. The database module stores compressed talents of community site visitors, facilitating retraining and checking out for the type module whilst bearing in mind recovery of proper internet web page traffic for submit-occasion evaluation and forensic functions. Evaluation of the framework become executed using the CICIDS2017 dataset, which reflects real network visitors; consequences display that the proposed framework achieves advanced accuracy in every binary and multiclass classifications compared to previous work, and excessive-level accuracy for restored website visitors. Furthermore, the framework’s modular layout enhances flexibility, taking into consideration smooth variation to specific network environments and evolving attack vectors. The integration of remarks mechanisms guarantees continuous improvement of the detection machine, allowing it to adapt to new threats. Finally, the potential software of this framework in side and fog networks is discussed, highlighting its relevance in the context of emerging technologies and disbursed computing environments.