Handling partially labeled network data: A semi-supervised approach using stacked sparse autoencoder

Network traffic analytics has become a crucial task in order to better understand and manage network resources, especially the softwarization era where implementation of this concept can be done easily with function virtualization. Currently, many approaches have been proposed improve performance classification. However, as new types emerge every day (and they are generally not labeled), opens challenge handled. Moreover, question how accurately classify using limited amount labeled data or partially is also another important concern. In fact, labeling often difficult time-consuming. solve previously described issues, we reformulate classification into semi-supervised learning both supervised (using data) unsupervised (no label combined. To do so, paper presents stacked sparse autoencoder (SSAE) based deep-learning model for The main motivations approach are: (i) unlabeled abundant available; (ii) whole greatly improved when large included training process; (iii) there limit much human effort thrown at problem. investigate our approach, an empirical study conducted on real dataset results indicate that SSAE pre-trained phase significantly model. Furthermore, compared against other representative machine-learning models, which Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Multi-Layer Perceptron (MLP), eXtreme Gradient Boosting (XGBoost), Autoencoder.