Linear support vector regression with linear constraints

This paper studies the addition of linear constraints to Support Vector Regression when kernel is linear. Adding those into problem allows add prior knowledge on estimator obtained, such as finding positive vector, probability vector or monotone data. We prove that related optimization stays a semi-definite quadratic problem. also propose generalization Sequential Minimal Optimization algorithm for solving with and its convergence. show an efficient this iterative closed-form updates can be used obtain solution underlying Then, practical performances are shown simulated real datasets different settings: non negative regression, regression onto simplex biomedical data isotonic weather forecast. These experiments usefulness in comparison more classical approaches.