Extended robust support vector machine based on financial risk minimization.

Area of Science:

• Machine Learning
• Financial Risk Management
• Statistical Modeling

Background:

• Support vector machines (SVMs) are increasingly used in machine learning.
• ν-support vector machine (ν-SVM) utilizes conditional value at risk (CVaR) for classification, a measure popular in finance for its subadditivity but sensitive to outliers.
• Existing robust SVM methods use truncated hinge loss, but there is a need for methods less sensitive to extreme values.

Purpose of the Study:

• To propose a novel classification method, extended robust SVM (ER-SVM).
• To develop a method that is less sensitive to outliers in the data distribution compared to ν-SVM.
• To introduce a risk measure that lies between CVaR and Value at Risk (VaR).

Main Methods:

• Proposed extended robust SVM (ER-SVM) classification method.
• Minimization of an intermediate risk measure between CVaR and VaR.
• ER-SVM is conceptualized as an extension of robust SVM utilizing a truncated hinge loss function.

Main Results:

• Numerical experiments suggest ER-SVM's potential for enhanced prediction performance.
• The proposed method demonstrates reduced sensitivity to outliers compared to ν-SVM.
• Optimal parameter selection is crucial for achieving superior prediction accuracy with ER-SVM.

Conclusions:

• ER-SVM presents a promising alternative for classification tasks sensitive to data outliers.
• The method offers a balance between risk management and predictive accuracy in machine learning.
• Further research and parameter tuning are recommended for practical applications of ER-SVM.