Joint label-specific features and label correlation for multi-label learning with missing label
- PDF / 1,434,527 Bytes
- 21 Pages / 595.276 x 790.866 pts Page_size
- 52 Downloads / 205 Views
Joint label-specific features and label correlation for multi-label learning with missing label Ziwei Cheng 1 & Ziwei Zeng 1
# Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Existing multi-label learning classification algorithms ignore that class labels may be determined by some features in the original feature space. And only a partial label of each instance can be obtained for some real applications. Therefore, we propose a novel algorithm named joint Label-Specific features and Label Correlation for multi-label learning with Missing Label (LSLC-ML) and its optimized version to solve the above-mentioned problems. First, a missing label can be recovered by the learned positive and negative label correlations from the incomplete training data sets, then the labelspecific features can be selected, finally the multi-label classification task can be modeled by combining the labelspecific feature selections, missing labels and positive and negative label correlations. The experimental results show that our algorithm LSLC-ML has strong competitiveness compared with some state-of-the-art algorithms in evaluation matrices when tested on benchmark multi-label data sets. Keywords Missing labels . Label-specific features selections . Positive label correlations . Negative label correlations
1 Introduction As an important branch of Artificial Intelligence, machine learning has been widely used in many practical applications, especially in the field of smart applied approaches. Many scholars have proposed many algorithms one after another, such as Zhou et al. [1], Bencherif et al. [2], Souza et al. [3], and Bae et al. [4]. These algorithms mainly use fuzzy neural networks to solve related problems. Traditional supervised learning methods are often only aimed at a single label, but in reality, an object often has multiple classification labels. This type of supervised learning is called multi-label classification. At present, multi-label learning has been widely used in web page classification [5], image annotation [6], biological analysis [7] and other fields. There are two ways to solve multi-label classification algorithm: problem transformation and algorithm
* Ziwei Zeng [email protected] Ziwei Cheng [email protected] 1
University of Science and Technology Liaoning, Anshan, China
adaptation. The idea of problem transformation is to transform multi-label classification into a series of single-label classifications, such as MLSVM [6] (Multi-label Support Vector Mechine). However, MLSVM does not take the label correlations into account. Different from the problem transformation, the idea of algorithm adaptation is to adapt the existing single-label classification algorithm to solve the multi-label classification, such as ML-KNN [8](Multi-Label K Nearest Neighbor), it introduces MAP(Maximum A Posteriori Probability Estimate) based on the traditional KNN algorithm, but the ML-KNN ignores the label correlations. Therefore, many algorithms related to label correlations have been proposed one after
Data Loading...
