Hyperparameter Selection in Support Vector Regression Using Difference of Convex Algorithms-Based Methods

Support Vector Regression (SVR) operates as an effective regression methodology within the machine learning landscape, with its performance largely dependent on hyperparameter tuning. This study proposes a bilevel optimization schema tailored for hyperparameter selection in SVR, subsequently recasting the problem into a Difference of Convex functions model (DC). We delineate convergence outcomes for this approach. Through computational experiments, the efficacy of this methodology is demonstrated, surpassing traditional hyperparameter selection techniques.