HSNN: A Hybrid Sampling With Neural Network Method for Imbalanced Communication Services Classification

Unlike traditional network traffic classification, communication service classification encompasses a broader spectrum of applications. Variations in individual user behavior over time lead to significant differences in the network traffic generated by various services. Consequently, in real-world scenarios, communication service datasets often exhibit severe class imbalance, with some categories containing substantially more data than others. To address these challenges, this paper proposes a novel hybrid sampling with neural network method (HSNN). The proposed approach integrates advanced sampling techniques, neural network architectures, and machine learning classifiers. Specifically, we preprocess the data using a time-windowing strategy, employ the Synthetic Minority Oversampling Technique (SMOTE) to mitigate data imbalance, and apply the Edited Nearest Neighbor (ENN) method to eliminate noisy samples. For classification, we combine an enhanced convolutional neural network with a random forest classifier. Experimental evaluations on our communication service dataset demonstrate that the HSNN framework achieves superior performance, highlighting its effectiveness in tackling the challenges inherent in imbalanced communication service data.