Integrated learning method and device based on interval optimization

Abstract

The invention discloses an integrated learning method and device based on interval optimization. The method comprises: step one, one group of trained base classifiers is obtained; step two, predictionvalues of all training samples by the base classifiers are obtained; step three, a weight of each base classifier is optimized by optimized integrated model interval distribution; and step four, weighted average processing is carried out on the prediction values of the samples by using the optimized weights to obtain a final prediction value and mark. In addition, the invention also discloses anintegrated learning device based on interval optimization. According to the invention, dependence on the specific base classification algorithm is avoided; and optimization is carried out based on theclassification error or AUC. Moreover, the whole model is easy to solve; and the applicability is high. During weight optimization, the accuracy and diversity are balanced by introducing the interval, so that the overfitting problem is solved and the prediction effect of the final integrated model is improved.

Description

technical field [0001] The invention belongs to the field of integrated learning in machine learning, and in particular relates to a weight optimization method and device for base classifiers in integrated learning. Background technique [0002] Integrated Learning: [0003] Among many machine learning methods, ensemble learning is the most successful type of method, among which methods such as random forest and gradient boosting are typical representatives of ensemble learning. The basic idea of ​​ensemble learning is to combine the results of many basic machine learning models to achieve the purpose of improving the learning effect. [0004] The two core steps of ensemble learning methods are the generation and integration of basic learning models. [0005] The generation of the model needs to complete the training task of the base classifier. For the input training data set, several different machine learning models are trained, and each of these models can predict the...

AI Technical Summary

Problems solved by technology

[0027] Although both diversity and spacing are considered to be key to the effectiveness of ensemble learning, the connection between the two has not been fully elucidated

Images