Robust classification via clipping-based kernel recursive least lncosh of error
作者：
Highlights：
• A novel robust classifier based kernel recursive least lncosh was proposed.
• The clipping concept is used to ignore the effect of large noises.
• Instead of the least square loss function, we provided the lncosh loss function.
• Convergence and stability were proved theoretically.
• Better classification on 500px social media data with real outliers was obtained.
摘要
•A novel robust classifier based kernel recursive least lncosh was proposed.•The clipping concept is used to ignore the effect of large noises.•Instead of the least square loss function, we provided the lncosh loss function.•Convergence and stability were proved theoretically.•Better classification on 500px social media data with real outliers was obtained.
论文关键词：Robust classification,Kernel recursive least lncosh,Non-Gaussian noise,Performance analysis,500px images,Social media data
论文评审过程：Received 19 August 2021, Revised 1 March 2022, Accepted 1 March 2022, Available online 15 March 2022, Version of Record 22 March 2022.
论文官网地址：https://doi.org/10.1016/j.eswa.2022.116811
