A study of PSVM and CPSVM models: analysis, correction, and application in operations research

Support Vector Machines (SVMs) are powerful tools in machine learning, widely used for classification and regression tasks. Recently, various extensions, such as Probability Estimation SVM (PSVM) and Conditional Probability SVM (CPSVM), have been proposed to enhance SVM performance across different conditions and datasets. In this article, we analyze these two models and provide insights into their behavior. A key contribution of this work is the derivation of necessary and sufficient conditions to guarantee the uniqueness of the solution in PSVM and CPSVM models. This theoretical aspect has not been previously addressed in the literature. Additionally, we identify errors in the dual formulation of CPSVM, propose a corrected version, and analyze its computational implementation. To better position these models within the framework of Operations Research (OR), we demonstrate their role in predictive analytics and optimization-based decision-making. Our study shows that CPSVM and PSVM approaches can be leveraged to enhance classification tasks in OR applications, such as healthcare, where accurate probability estimation is critical. Our empirical evaluation, conducted on diverse benchmark datasets, including an application to Fish Schools classification, demonstrates that CPSVM models outperform traditional SVM approaches, especially in nonlinear settings. We further validate these improvements through statistical tests, including the Friedman test and post hoc analysis, confirming that CPSVM offers statistically significant advantages in probabilistic classification. These results underscore the relevance of our corrected CPSVM formulation for OR applications requiring robust predictive models.