Zeekflow+: A Deep LSTM Autoencoder with Integrated Random Forest Classifier for Binary and Multi-class Classification in Network Traffic Data
Emmanouil Arapidis, Nikos Temenos, Dimitrios Giagkos, Ioannis Rallis, Dimitris Kalogeras, Nikolaos K. Papadakis, Αντώνιος Λίτκε, Sotirios Messinis · 2024
This work proposes Zeekflow+, a Deep LSTM Autoencoder (AE) architecture with integrated Random Forest (RF) classifier for effective binary & multi-class classification of network traffic data. The Deep LSTM AE is used to extract underlying patterns existing within the features of the input data and encode them accordingly, allowing for binary classification to malicious or benign behavior. Through the use of the RF classifier, the binary classification capabilities are extended to multi-class ones for effectively identifying the origin of the attacks. This is achieved by training the RF classifier using the total reconstruction loss and the Deep LSTM AE encoded data. Experimental results on the USTC-TFC2016 dataset, showcase the performance of the Zeekflow+ architecture in multi-class classification resulting in more than 99% in the precision, recall and F1-Score classification metrics. To further demonstrate the effectiveness of the Zeekflow+ architecture, it is used for binary classification task in the same dataset, while considering different Floating Point arithmetic quantizations with the results showing negligible performance drop, making it suitable for real-time IoT edge device deployment.