PDF Publication Title:
Text from PDF Page: 107
HMM, and ARIMAX (see Figure 7). 3.3 Other results In average, the devices are active 58% of the time (this can be seen as well at the bottom chart in Figure ). That implies that our baseline accuracy is 0.58 as this value can be obtained simply by giving a constant value of ‘on’ activity for all devices and times. In order to improve the final prediction, we tried to ensemble [10] the results from the different methods using a voting schema to choose the final one, and, in all cases, the result obtained was worse than the result from the best algorithm used in the input mix. This bad result from ensembling, suggests again that there is not a variance problem and that prediction improvement will come from an improvement at the bias side. It is interesting to compare the computing requirements for the training of the different methods. In Figure 7 we provide the results; the values correspond to training times for 6214 SIMs for a forecasting interval of 48 hours. The complete process was done 4 times for consecutives intervals of 48 hours, in order to have significant data to compute the average of the forecasting values. In the figure, training times are given in a log scale due to the big difference between values. We can see that the non-time-series methods requires in general less computing time, and that ARIMAX and HMM require much more computing time that the rest. Figure 7. Computing time used to train the different methods. 3.4 Considerations about the forecasting interval The forecasting interval used for the study has been a 1-hour interval, performing all forecasting calculations over the next 48 1-hours. A 1-hour interval value has been also used to aggregate the on/off activity for each SIM for the training data. The 1-hour period used both as the training feature (activity) to perform predictions, and, as Doctoral Thesis: Novel applications of Machine Learning to NTAP - 105PDF Image | Novel applications of Machine Learning to Network Traffic Analysis
PDF Search Title:
Novel applications of Machine Learning to Network Traffic AnalysisOriginal File Name Searched:
456453_1175348.pdfDIY 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 | RSS | AMP |