Cluster-based multi-label imbalance biomedical data classification method

Abstract

The invention relates to a cluster-based multi-label imbalance biomedical data classification method. The method includes the following steps of S101 defining a relation matrix for the label imbalance data according to the feature similarity and label relevance; S102 clustering the data according to the relation matrix; S103 directionally increasing the imbalance labels in each cluster; S104 training and learning the data in each cluster by means of a multi-label classifier; and S105 combining the result of each classifier according to the polling rule and predicting the label. The data is clustered by means of a hierarchical clustering method, the label relevance is considered during the clustering to reduce the imbalance of the labels in the clusters, so that the reliability for new data generation by means of a re-sampling method is improved, and the probability for noise data is reduced.

Description

technical field [0001] The invention relates to the field of multi-label technology, in particular to a clustering-based multi-label imbalance biomedical data classification method. Background technique [0002] Multi-label learning can be called a paradigm of supervised learning. Unlike binary classification problems, multi-label learning allows samples to have multiple categories. But also different from multi-classification problems, multi-label learning allows an object to belong to multiple categories at the same time. For the classification of multi-label sample data, it means that a certain sample data contains multiple labeled labels, the entire sample data set is marked into a q-dimensional multi-label space, and the feature vector of each data sample is expressed as x i , the label vector labeled d i ={d i1 , d i2 ,...,d iq}, where d ir ∈{0,1}, 1≤r≤q, means that for each sample data, the label has the same label, 1 indicates that the sample data contains the...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the object of the present invention is to provide a clustering-based multi-label imbalanced biomedical data classification method, aiming to solve the problem that most traditional classifiers have errors in multi-label imbalanced biomedical data classification performance

Images