Improving the Classification of Network Slice Using SMOTETomek and Machine Learning Models in 5G Networks

Network slicing (Net-Slice) is considered an essential feature of 5G networks. Net-slice is enabled to support a variety of developing applications with increased flexibility and performance criteria by dividing the physical network infrastructure into multiple logical networks. As a result of these applications, a huge amount of data has been generated across a large number of cellular devices, and it has to be efficiently used. In this context, this paper aims to work on the available 5G-enabled net-slice dataset and apply different Machine learning (ML) models for the classification of slice types. The 5G Net-Slice dataset contains the data generated from various devices, such as smartphones. First, we used the SMOTETomek (Synthetic Minority Oversampling Technique Tomek) method to handle the class imbalance problem. To classify slice types, this study considers various ML models, namely AdaBoost, SVM (Support Vector Machine), Extra Tree, k-NN (k-Nearest Neighbors), xGB (extreme gradient boosting), RF, Bagging, Histgardient boosting (HistGB), LGBM (Light gradient boosting machine), GB (Gradient boosting), Stacking, and MLP (Multi-layer perceptron). Furthermore, we applied various ML models to classify the slice type and then compared the performance of all the models. We evaluated these models on the accuracy, recall, precision, and Fl-Score performance matrices. Results demonstrate that Extra Tree outperforms other models with SMOTETomek in the classification of net slices in 5G.