Search

In the classification problem, the commonly algorithms are semi-supervised learning algorithm, Bagging and Boosting classification algorithm. When there is little mark data and the difference between data is large, it's hard to find effective regulations to classify them. This paper proposed a Boosting classification algorithm under triple set constraints to solve this problem. The algorithm divided the mark data, pseudo mark data and unmarked data by triple constraints, and introduced a balance function to weight the two neighbors of the mark data and established the stable data for this space. Then it iterated the classifier according to that data information. Through the gradient descent method, the balance function converged and finally got the pseudo-marker data and classifier. Experiments on nine datasets of UCI verify that this algorithm is more efficient and feasible. [ABSTRACT FROM AUTHOR]