Novel applications of Machine Learning to Network Traffic Analysis

PDF Publication Title:

Novel applications of Machine Learning to Network Traffic Analysis ( novel-applications-machine-learning-network-traffic-analysis )

Previous Page View | Next Page View | Return to Search List

Text from PDF Page: 125

with the other models. C. Impact of time-series length An important parameter to be studied is the influence of the number of packets to be considered when we analyze the network flows. There are flows with hundreds of packets whereas others have only a single packet. An important doubt at the beginning of the study was the possible influence of this parameter, since increasing the number of included packets could improve detection but at the cost of much higher computing time and resources. As a balanced decision, we opted for a maximum of 20 packets. We have considered only the first 20 packets exchanged in a flow lifetime. In the case of flows with more than 20 packets, we have disregarded any packet after packet number 20. Flows with less than 20 packets were padded with zeros. Then it was important to know the impact of the number of packets in the overall detection problem and to confirm whether 20 packets was a sensible number. To this aim, we analyzed the performance considering a different number of packets and different architectures. We present here the results for two representative architectures (RNN-1 and CNN+RNN-2a) We can see the results for the RNN-1 architecture in Fig. 10, and it is important to note that overall detection quality is not significantly changed by using fewer packets until the number of packets is less than five per flow. Therefore, it is enough to consider the very first packets of a flow to have most of the information that allows us to infer their service. In Fig. 10 we can easily appreciate that the detection quality starts to degrade when the number of packets per flow is less than five. Fig. 10. Classification performance metrics (aggregated) vs. time-series length for architecture RNN-1 Fig. 11 shows the results on the impact of the number of packets for the architecture CNN- RNN-2a. This model supports a smaller reduction in the number of packets (the model needs a minimum length of 7 in the temporal-dimension), but we can still appreciate that regardless of an initial performance decrease, this reduction is not monotonic with the decrease in the number of packets, in fact, it keeps approximately constant for packets lengths in the interval from 7 to 15 (disregarding some intermediate noisy values). Doctoral Thesis: Novel applications of Machine Learning to NTAP - 123

PDF Image | Novel applications of Machine Learning to Network Traffic Analysis

PDF Search Title:

Novel applications of Machine Learning to Network Traffic Analysis

Original File Name Searched:

456453_1175348.pdf

DIY PDF Search: Google It | Yahoo | Bing

Cruise Ship Reviews | Luxury Resort | Jet | Yacht | and Travel Tech More Info

Cruising Review Topics and Articles More Info

Software based on Filemaker for the travel industry More Info

The Burgenstock Resort: Reviews on CruisingReview website... More Info

Resort Reviews: World Class resorts... More Info

The Riffelalp Resort: Reviews on CruisingReview website... More Info

CONTACT TEL: 608-238-6001 Email: greg@cruisingreview.com (Standard Web Page)