Evaluation of Bernstein Polynomial as a Machine Learning technique

Abstract

Abstract Regression algorithms are used to device a function to predict the output based on that function for new unseen inputs. We have used Bernstein polynomial as a type of approximation curve to fit a polynomial between all features of a dataset so as to perform regression on it. As approximation curves do not generally meet most of the input points, which are used to create it, hence the chances of it over-fitting to a model are low. In this paper, we have used Bernstein polynomial separately for each feature versus output graph, assuming that each feature in the dataset is independent of each other and combines the outcome of each curve to predict the final result.