Semi-supervised Weighted Ternary Decision Structure for Multi-category Classification
- PDF / 1,691,922 Bytes
- 28 Pages / 439.37 x 666.142 pts Page_size
- 73 Downloads / 184 Views
Semi-supervised Weighted Ternary Decision Structure for Multi-category Classification Pooja Saigal1 · Reshma Rastogi2 · Suresh Chandra3
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Semi-supervised learning has attracted researchers due to its advantages over supervised learning. In this paper, an extremely fast multi-category classification algorithm, termed as weighted ternary decision structure (WTDS) is proposed. WTDS is a generic algorithm that can extend any binary classifier into multi-category framework. This work also proposes a novel semi-supervised binary classifier termed as Weighted Laplacian least-squares twin support vector machine which is further extended using WTDS. The novel semi-supervised classifier obtains the solution by formulating a pair of Unconstrained Minimization Problems which are solved as systems of linear equation. WTDS takes advantage of the strengths of the classifier and efficiently constructs the multi-category classifier model in the form of a decision structure. WTDS outperforms other state-of-the-art multi-category approaches in terms of classification accuracy and time complexity. To confirm the feasibility and efficacy of proposed algorithm, experiments are conducted on benchmark UCI datasets. Keywords Semi-supervised learning · Multi-category classification · Weighted ternary decision structure
1 Introduction In many practical problems, the cost of obtaining the labels is very high as compared to the cost of generating the data, for example speech recognition [32], web page classification [9], video surveillance [8] etc. Due to insufficient volume of labeled data, the traditional supervised learning methods may not be able to accomplish the desired objective. Therefore,
B
Pooja Saigal [email protected]; [email protected] Reshma Rastogi [email protected] Suresh Chandra [email protected]
1
Vivekananda Institute of Professional Studies, New Delhi, India
2
Department of Computer Science, South Asian University, New Delhi, India
3
Ex Faculty, Department of Mathematics, Indian Institute of Technology, New Delhi, India
123
P. Saigal et al. Table 1 Popular semi-supervised approaches Generative model
[14]
An extension of semi-supervised mixtures to a multi-view setting and mixtures of factors analyzers. The model is old and simple, but it assumes a certain type of distribution for unlabelled data
Self-learning model
[21]
Assigns hard labels to unlabeled samples. This model considers a very few labeled samples, and therefore suffers from poor prediction
Co-training model
[35]
Uses few labels but prediction is not good
Transductive learning model
[15,31]
Model is less susceptible to over-fitting as compared to self-learning and co-training models. But the prediction is classifier dependent
Graph-based learning model
[1]
Graph-based learning has better classification accuracy, but with high computational complexity. These are widely used for pattern recognition
the semi-supervised learning (SSL
Data Loading...
