Feature Selection Of Data Gravitation Learning Model And Its Application In Video Traffic Recognition

The rapid development of network applications has simulated the growth of network traffic.Especially in recent years,the explosive growth of video traffic is closely related to video applications.How to identify and supervise the different types of video traffic has become an urgent topic for both government and Internet companies.The feature selection from many aspects including time,space,package level,flow level,etc.generates a large amount of feature data,which brings about the high cost of storage as well as the low recognition efficiency.Categorizing the features of the original video traffic,and selecting those that improve the efficiency and accuracy of the classification,has become a new path to solve the problem.Besides,as many other practical classification problem,identification of the video traffic with imbalance is another crucial problem.To address the above-mentioned key problems in network video traffic identification,this paper introduces data gravitation classification model in data recognition,and feature selection algorithm in data processing.Through the study of publicly imbalanced data sets and the collection of video traffic data,the research work has been systematically carried out as follow.(1)Aiming at the classification of imbalanced data,the gradient descent algorithm is used to optimize the data gravitation classification model,and a new data gravitation classification model based on gradient descent(IDGC-GD)is proposed.This method defines the loss function as the gravitational difference between categories,iteratively updates the feature weights of the data gravitation classification model through the gradient descent algorithm,using a non-linear decreasing strategy to improve the convergence speed of the algorithm.The algorithm has been experimented with 7 existing different types of imbalanced learning methods on 21 public imbalanced data sets,which proves that the model has significant advantages in imbalanced data classification.(2)Aiming at the feature selection problem of imbalanced data,the study proposed a probability density function method for precise classification of features.This method uses the distribution distance to define the distribution difference between the features,and uses the discretization operation based on the sliding window to discretize the feature value for the continuously distributed features.Combining Wasserstein distance to calculate the weight,the study weights the data features,and finally uses the imbalanced data gravitation classification model(IDGC)for data classification to verify the effect of the feature selection method.This paper compares 4 different strategies through 30 public data sets,and finally selects the feature selection method with the best results.The experimental results show that this feature selection method can improve the classification accuracy and the efficiency of the algorithm.(3)Aiming at the identification of network video traffic,this research proposes a new feature selection technical framework,revolving around an accurate flow identification model developing from the imbalanced data gravitation classification model based the gradient descent(IDGC-GD),and a feature flow extraction method related to content.4different types of data sets are collected from real video websites for model testing.The experimental results show that this feature selection framework is effective in identifying video traffic in practical.In summary,this study proposes an effective solution to the feature selection of the data gravitation classification model and its application in video traffic recognition.The effectiveness of the method proposed in this article is verified through experiments.The research has certain theoretical significance and application value for advancing the optimization of data gravitation classification model and the identification of network video traffic.