RoBoSS: A Robust, Bounded, Sparse, and Smooth Loss Function for Supervised Learning.

Area of Science:

• Machine Learning
• Supervised Learning
• Data Science

Background:

• Traditional loss functions in machine learning struggle with outlier-prone and high-dimensional data.
• This limitation leads to suboptimal outcomes and slow convergence in supervised learning algorithms.

Purpose of the Study:

• To propose a novel robust, bounded, sparse, and smooth (RoBoSS) loss function for supervised learning.
• To integrate the RoBoSS loss into a Support Vector Machine (SVM) framework, creating a new robust algorithm named RoBoSS-SVM.
• To theoretically analyze the classification-calibrated property and generalization ability of the proposed method.

Main Methods:

• Development of the RoBoSS loss function.
• Integration of RoBoSS loss with SVM to create the RoBoSS-SVM algorithm.
• Empirical validation on 88 benchmark datasets from KEEL and UCI repositories, including datasets with added outliers and label noise.
• Evaluation on biomedical datasets (EEG and breast cancer).

Main Results:

• The RoBoSS-SVM algorithm demonstrated superior generalization performance compared to existing methods.
• The proposed algorithm showed significant efficiency in training time.
• Validation confirmed the robustness of the RoBoSS loss function and RoBoSS-SVM in handling challenging datasets with outliers and noise.

Conclusions:

• The novel RoBoSS loss function and RoBoSS-SVM algorithm effectively address limitations of traditional loss functions.
• The RoBoSS-SVM model offers a robust and efficient solution for supervised learning tasks, particularly with noisy and high-dimensional data.
• The algorithm shows promise for applications in machine learning and the biomedical domain.