A study on application of semi-supervised collaborative classification algorithm

Abstract

The treatment method of Tri-Training algorithm in classifier selection and confidence estimation breaks through the limitation of Co-training algorithm. In order to further improve the classifiers' performance, a semi-supervised collaborative classification algorithm with enhanced difference makes some improvement respectively on classifier diversity, model update strategy and unlabeled sample prediction method. Because of the use of different classifiers and consideration of classifier diversity, this algorithm has good performance in unbalanced sample set classification. Establish classification model based on the above algorithm, and use it to do experiment with bridge structural health monitoring data, the results of which demonstrate the validity and applicability.