Multi-class SVM research is ongoing research, but methods modeled for precise data can be less accurate due to measurement, and modeling errors. In such a situation, we face a set of uncertain data sets. This paper introduces a multi-class SVM using regular simplex for stochastic inputs. The presented model is based on a regular simplex support vector machine with probabilistic constraints, which is investigated in two states of known and unknown populations. We put a noise in the constraint which comes up with a known distribution. We release the constraints from the probabilistic state using statistical theories and moment estimation. In the simulation part, we use multinomial logistic regression to fit the relationship between features and labels through the Monte Carlo method. We demonstrate that the proposed model is more efficient than the model relying on accurate data. This paper presents an improved version of the RSSVM model.

Keywords