A Comparative Study on Machine Learning-Aided Flow Traffic Generators

In recent years, machine-learning algorithms have been developed for application-level traffic generators, but have yet to be implemented in flow-level traffic generators. Since flow-level traffic data is a time series traffic data, it could be used for predicting subsequent network traffic. Its prediction ability can be utilized to generate traffic when the database cannot provide the data. This is an improvement from our previous work that uses exact-database methods which only generate traffic data based on the database. A comparison is made between traditional machine-learning methods (Linear Regression, Support Vector Regression, and Gaussian Process Regression) and a deep-learning method (Long-Short Term Memory) to determine how well the Artificial intelligence (AI) algorithm performs when applied to a flow-level traffic generator. These methods enable traffic generators to continue generating traffic while the database is unavailable. Among all the methods, the Long-Short Term Memory (LSTM) algorithm performs the best because it has the lowest Root Mean Square Error (RMSE) value which is 0.13. Visually, some traditional machine learning methods still manage to imitate real traffic, despite their high RMSE values.