PDF Publication Title:
Text from PDF Page: 158
In this section, we provide the performance metrics obtained for all the models analyzed for this work. The models are described in a previous section (Section III.C). The performance metrics for all the models are presented in Fig. 5. The performance metrics in Fig. 5 are aggregated metrics using a weighted average. In Fig. 5, we can see that Models 2 and 3 present the best results in terms of F1-score. Of these models, the best accuracy is given in Model 2 (0.6218) and the best F1-score is obtained for Model 3 (0.6987). The slightly better result of Model 3 has to be balanced with the greater needs of memory and processing time for this model. We see that, in general, the metrics obtained are not high, but these metrics are formed by adding the results of 14 labels. In addition, there are three highly unbalanced labels (blur, blockness and black pixel) that significantly worsen the results. This can be seen in Fig. 6, where the performance metrics are calculated for each label separately. Considering the other models, Random Forest offers fairly good results but the final F1- score (0.6787) is below the best models due to poor results in the recall metric. Focusing on the F1-score, which is our preferred metric for aggregated results (Fig. 5), Model 3 provides a 3% increase over Random Forest. To calculate AUC scores it is necessary to obtain prediction probabilities, which can be cumbersome and not accurate for some models. For example, SVC requires an additional Platt scaling or some other alternative method to retrieve prediction probabilities. Therefore, the AUC score is not considered in Fig. 5 for aggregated results for all models. It is important to highlight that the worst result in Fig. 5 is for Model 1. This model is formed exclusively by an RNN network and does not include a CNN in its architecture. This is a rather unexpected result as the inclusion of a CNN was not anticipated as something so critical, considering the time-series nature of the predictors. Fig. 5. Performance metrics (aggregated) for QoE classification for all models In Fig. 6 is provided the one-by-one results for the classification of the 7 labels for the current and next time-steps. The AUC score and the frequency of the non-error value are given for all the labels, in addition to the metrics: accuracy, F1, precision and recovery. The results are obtained with Model 2. It is important to note in Fig. 6 that for some labels the results are very good (chrominance Doctoral Thesis: Novel applications of Machine Learning to NTAP - 156PDF 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 (Standard Web Page)