Data flow classification method and device based on dynamic fast decision tree algorithm

Abstract

The embodiment of the invention discloses a data flow classification method and device based on a dynamic fast decision tree algorithm. According to the method, a difference value between a heuristicmetric value of a first optimal attribute and a heuristic metric value of an empty attribute is compared with a first current splitting threshold value, and the heuristic metric value of the empty attribute is smaller than a second optimal attribute, so that the difference value between the heuristic metric value of the first optimal attribute and the heuristic metric value of the empty attributeis greater than the first current splitting threshold value; the splitting speed of the fast decision tree is much higher than that under a conservative mechanism, the learning ability of the online decision tree can be brought into full play, and for the internal node, the difference value between the heuristic metric value of the second optimal attribute and the heuristic metric value of the splitting attribute of the node is compared with a second current splitting threshold value. When a judgment is made that the original splitting attribute for splitting is not the optimal attribute, nodes are spitted by using the optimal second optimal attribute so as to quickly adapt to concept drift and improve the algorithm accuracy.

Description

technical field [0001] The invention relates to the technical field of decision trees, in particular to a method and device for classifying data streams based on a dynamic fast decision tree algorithm. Background technique [0002] The online decision tree is one of the most popular algorithms in the data stream classification algorithm. During the learning process, the online decision tree algorithm will frequently perform split attempts based on the selected best split attributes to ensure that the leaf nodes split in time to achieve data stream classification. [0003] In a non-stationary environment, the distribution of data may change over time, that is, concept drift. When concept drift occurs, the optimal split attribute of the leaf node will usually change drastically, and it is difficult in this case Choosing an appropriate split attribute for splitting makes it difficult for the online decision tree algorithm using a conservative mechanism to quickly recover from ...

AI Technical Summary

Problems solved by technology

[0003] In a non-stationary environment, the distribution of data may change over time, that is, concept drift. When concept drift occurs, the optimal split attribute of the leaf node will usually change drastically, and it is difficult in this case Choosing an appropriate split attribute for splitting makes it difficult for the online decision tree algorithm using a conservative mechanism to quickly recover from drift, so this conservative mechanism limits the learning ability of the online decision tree and cannot quickly adapt to concept drift, resulting in Algorithms are less accurate

Images