Multi-classification general integration method of adaptive weight

Abstract

The invention provides a multi-classification general integration method of adaptive weight. Different from a common ensemble classification method only aiming at one classification task, the method comprises the following steps: firstly, determining a weight allocation coefficient for different classification problems, and then, automatically calculating a weight of primary model fusion according to a model evaluation index and the weight allocation coefficient by using an exponential function distribution characteristic; finally, the fusion weight is adaptively adjusted through a continuous iteration method, and model fusion under different classification tasks is achieved. Experiments show that the method can realize model fusion on nine data sets with different classification numbers, different fields and different scales, and the fusion effect of the method is superior to that of a baseline method in most tasks. Compared with a single model, a result F1 value fused by the method provided by the invention is stably increased by 3-25 percentage points; compared with other integration methods, the method has the advantages that the F1 value is stably increased by 1-2%, and universality and effectiveness of the method are proved.

Description

technical field [0001] The invention relates to the technical field of data mining and data classification processing, in particular to a multi-class general integration method with self-adaptive weight. Background technique [0002] The main idea of ​​ensemble learning is group decision-making, which fuses the results of multiple weak learners through various combination mechanisms to obtain better results than any single composition algorithm, and to improve accuracy and stability. [0003] In the current research work, most of the methods focus on adjusting the weights in the integration process to obtain good classification results. In general, the existing research work can be divided into methods during the training process and methods after training. Among them, the research on implementing fusion in the training process can be traced back to 1990. Freund and Shapire proposed the Boosting algorithm. By training multiple weak classifiers, each weak classifier is given ...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that existing integration methods cannot adaptively select integration weights when facing multiple classification problems, and have no versatility, the present invention provides a multi-category general integration method with adaptive weights

Images