Abstract
With a rapid evolution of technology, threats on networked systems have become increasingly pervasive, and detection approaches need complex and advanced methods to analyze network activities to respond effectively. This research explores the effectiveness of various methods such as Deep Neural Networks, Support Vector Machine (SVM), Decision Tree, and Random Forest classifiers for detecting and predicting network threats and therefore develops a model evaluation pipeline. In the study, we used ACI-IOT-2023 dataset, including packet headers and payloads, which contain diverse instances of network traffic and potential threat data. The study evaluates the models using statistical performance metrics and cross validation techniques to ensure robustness of the model. The study also explores the feature sets that can be selected to improve the speed of the system and get the highest detection rates out of the machine learning algorithms, considering the high number of features available and open nature of payloads. The research highlights the importance of selecting an appropriate model for enhancing the accuracy and efficiency of threat detection systems. With that, we leave room for further exploration of models including neural networks to further optimize performance, and to improve the detection capabilities and efficiency of the system.