ELM-MC: multi-label classification framework based on extreme learning machine
- PDF / 2,046,552 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 16 Downloads / 233 Views
ORIGINAL ARTICLE
ELM‑MC: multi‑label classification framework based on extreme learning machine Haigang Zhang1 · Jinfeng Yang1 · Guimin Jia2 · Shaocheng Han3 · Xinran Zhou4 Received: 27 December 2018 / Accepted: 2 March 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract Multi-label classification methods aim to a class of application problems where each individual contains a single instance while associates with a set of labels simultaneously. In this paper, we formulate a novel multi-label classification method based on extreme learning machine framework, named ELM-MC algorithm. The essence of ELM-MC algorithm is to convert the multi-label classification problem into some single-label classifications, and fully considers the relationship among different labels. After the classification of one label, the associations with next label are applied to update the learning parameters in ELM-MC algorithm. In addition, we design a backup pool for the hidden nodes. It can help to select relatively suitable hidden nodes to the corresponding label classification case. In the simulation part, six famous databases are applied to demonstrate the satisfied classification accuracy of the proposed method. Keywords Multi-label classification · Extreme learning machine · Principle component analysis · Linear discriminant analysis
1 Introduction Supervised learning belongs to the traditional machine learning task of determining a function from labelled training data. The classification task, as one of supervised learning fields, aims to map the instances to discrete labels. The multi-label classification is concerned with learning from a set of individuals that are associated with more than one label. With the arrival of data age, the demand of multi-label classification applications is growing, such as music categorization, text categorization and scene classification etc. [1]. The main challenge of multi-label classification task is the overwhelming size of output label space. With the increase * Jinfeng Yang [email protected] 1
Institute of Applied Artificial Intelligence of the Guangdong‑Hong Kong‑Macao Greater Bay Area, Shenzhen Polytechnic, Shenzhen 518055, China
2
Tianjin Key Laboratory for Advanced Signal Processing, Civil Aviation University of China, Tianjin 300300, China
3
Basic Experimental Center, Civil Aviation University of China, Tianjin 300300, China
4
School of Computer Science and Engineering, Central South University, Changsha 410083, China
of the number of labels, the label space presents exponential expansion [2]. The standardized definition for multi-label classification problem can be summarized as • Let X ∈ Rd denotes a d-dimensional instance space with [ ] T
numberical or categorical features. Xi = xi1 , xi2 , … , xid represents the ith instance with the feature attributes of xi1 , xi2 , …{ , xid. } • Let L = l1 , l2 , … , lq represents the finite label space to which the samples belong. Necessarily, q > 0 . [ ]T Li = li1 , li2 , … , liq = [−1, 1]q denotes the
Data Loading...
