Multi-target regression method based on multi-class multi-label evolution super network

Abstract

The invention requests to protect a multi-target regression method based on a multi-class multi-label evolution super network. The method comprises the following steps: 101, converting a multi-target regression problem into a multi-class multi-label classification problem; 102, carrying out modeling on the multi-class multi-label data by adopting an evolution super-network model; 103, representing high-order correlation among a plurality of output targets by adopting an evolutionary super-network model; 104, fusing cost-sensitive learning and an evolution super-network model, and solving the problem of class imbalance; and 105, simultaneously predicting the output of a plurality of mutually associated targets by establishing a multi-category multi-label evolution super network model. The multi-target regression problem is converted into the multi-class multi-label classification problem, the incidence relation among multiple output targets in the multi-target regression problem is effectively represented, the multi-class multi-label evolution super-network model is trained, the output of the multiple correlated targets is predicted at the same time, and therefore, the prediction performance of multi-target regression is improved.

Description

technical field [0001] The invention belongs to the field of machine learning and artificial intelligence, and in particular relates to a multi-objective regression method based on multi-class multi-label evolution supernetwork. Background technique [0002] Traditional supervised learning is the most studied and widely used learning framework in the field of machine learning. In this framework, the learning system usually uses an attribute vector to describe the essence of the learning object, and at the same time uses a category label to represent the semantic information of the learning object, and a data sample passes an attribute vector and a category label associated with the vector To represent that a data sample corresponds to only one category label and has only one semantic meaning. Traditional supervised learning has achieved great success in dealing with learning objects with clear and single semantics. However, with the development of information technology, t...

AI Technical Summary

Problems solved by technology

[0006] (2) The multi-objective regression problem needs to deal with the mapping relationship between input and multiple output targets at the same time, and these mapping relationships may be linear or non-linear, and some may even be linear while the other part is non-linear

[0007] (3) The existing multi-objective regression methods are basically based on the same feature expression. Although these methods can use the correlation between multiple output targets, they ignore the particularity of each output target.

Images