A Parameter-Free Self-Training Algorithm for Dual Choice Strategy

Wei Zhao, Qingsheng Shang, Jikui Wang, Xiran Li, Xueyan Huang, Cuihong Zhang

2024

Abstract

In the field of machine learning, semi-supervised learning has become a research hotspot. Self-training algorithms, improve classification performance by iteratively adding selected high-confidence samples to the labeled sample set. However, existing methods often rely on parameter tuning for selecting high-confidence samples and fail to fully account for local neighborhood information and the information of labeled samples. To address these issues, this paper proposes a self-training algorithm with a parameter-free self-training algorithm for dual choice strategy. Firstly, the selection problem of K-value in KNN classifier is solved by using natural neighbors to capture the local information of each sample, and secondly, adaptive stable labels are defined to consider the information of labeled samples. On this basis, a decision tree classifier is introduced to combine the global information for double selection to further select high-confidence samples. We conducted experiments on 12 benchmark datasets and compared them with several self-training algorithms. The experimental results show that the FSTDC algorithm achieves significant improvement in classification accuracy.

Download


Paper Citation


in Harvard Style

Zhao W., Shang Q., Wang J., Li X., Huang X. and Zhang C. (2024). A Parameter-Free Self-Training Algorithm for Dual Choice Strategy. In Proceedings of the 1st International Conference on Data Mining, E-Learning, and Information Systems - Volume 1: DMEIS; ISBN 978-989-758-715-3, SciTePress, pages 102-107. DOI: 10.5220/0012920900004536


in Bibtex Style

@conference{dmeis24,
author={Wei Zhao and Qingsheng Shang and Jikui Wang and Xiran Li and Xueyan Huang and Cuihong Zhang},
title={A Parameter-Free Self-Training Algorithm for Dual Choice Strategy},
booktitle={Proceedings of the 1st International Conference on Data Mining, E-Learning, and Information Systems - Volume 1: DMEIS},
year={2024},
pages={102-107},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012920900004536},
isbn={978-989-758-715-3},
}


in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Data Mining, E-Learning, and Information Systems - Volume 1: DMEIS
TI - A Parameter-Free Self-Training Algorithm for Dual Choice Strategy
SN - 978-989-758-715-3
AU - Zhao W.
AU - Shang Q.
AU - Wang J.
AU - Li X.
AU - Huang X.
AU - Zhang C.
PY - 2024
SP - 102
EP - 107
DO - 10.5220/0012920900004536
PB - SciTePress