A student conducted a similar project called "Machine Learning in Traffic Classification of SDN" and created features based on "packet flows." However, if this method of feature extraction was taken, I would have to make a script to convert our packets into flows. The algorithm of choice would be logistic regression.
I also discovered this published paper called "Deep Learning for Encrypted Traffic Classification." This paper described the limitations of classical ML techniques for classifying encrypted network traffic. The paper then goes on to discuss the pros and cons of deep learning and even provides summaries of deep learning methods that were used in other published works.