Novel Bayesian weighting method based on CFS _ KL

Abstract

The invention discloses a novel Bayesian weighting method based on CFS _ KL, and the method comprises the steps: taking a fingerprint name in a fingerprint library as the mark data of a model, and forming training data through a response sequence flag bit under the fingerprint name; performing box sealing preprocessing operation on the training data; calculating the correlation degree between theattributes and the classes by using KL divergence to serve as the weight of each attribute; selecting 42 dimensions by using a feature selection method; correcting the weight calculated by the KL divergence by using the dimension selected by the CFS; performing training by using a weighted Bayesian algorithm; inputting the vector into a trained fingerprint model through box sealing operation, calculating the maximum posterior probability of each flow based on a CFS _ KL weighted Bayesian algorithm, and completing simulation data testing; collecting real traffic in a mode of sending a packet tothe target network segment, inputting the real traffic into the fingerprint model, and predicting a result; and calculating the test precision of the real flow. According to the method, the requirement of the Bayesian algorithm for feature independence is relieved, and the recognition precision of the Bayesian algorithm is improved.

Description

technical field [0001] The invention belongs to the technical field of machine learning, and in particular relates to a novel Bayesian weighting method based on CFS_KL (correlation-based feature selection_Kullback-Leibler). Background technique [0002] As one of the ten classic algorithms of machine learning, Bayesian algorithm has many applications in many fields, and all of them have shown very good results. For example, judging whether an email is spam or not based on its title and content. However, since learning the optimal Bayesian classifier is just like learning the Bayesian network, it is an NP-hard problem, so learning the Naive Bayesian classifier has been favored by many scholars, and Naive Bayesian is often based on a simple But unrealistic assumption: The features of the training data are independent of each other. This strong condition is difficult to achieve in real life. Even in reality, it has logically shown that the features are independent of each other...

AI Technical Summary

Problems solved by technology

However, since learning the optimal Bayesian classifier is just like learning the Bayesian network, it is an NP-hard problem, so learning the Naive Bayesian classifier has been favored by many scholars, and Naive Bayesian is often based on a simple But unrealistic assumption: the features of the training data are independent of each other. This strong condition is difficult to achieve in real life. Even in reality, it has logically shown that the features are independent of each other. In the actual data Not so independent, which greatly limits the classification effect of Naive Bayes

Images