Self-learning consensus voting strategy and its application based on multiple classification algorithms

Machine learning classification algorithms are widely used, and each algorithm has its own characteristics. It is not straightforward to determine which classification algorithm should be used in different circumstances and how to improve the accuracy. This paper proposes a self-learning consensus voting strategy based on a variety of classification algorithms to improve the accuracy of classification. First, the classification model is trained using historical records, then the classification result is obtained by a self-learning consensus voting strategy, and finally the best classification result is attained. Through such algorithm fusion and voting, the practical problem of classification is suitably addressed. Based on a handwritten numeral dataset, this paper tests, analyzes, and compares the nine most common classification algorithms in machine learning, and makes a self-learning consensus vote on the classification result, which is verified by an intelligent handwritten numeral marking system. The system has gone through the steps of image acquisition, document upload, image preprocessing, classification algorithm learning, voting, answer comparison and scoring, etc., for examination paper autocorrection. The experimental results show that the recognition accuracy can be improved by more than 2% through the self-learning consensus voting strategy, and this voting method is universal and can be applied in practice. © 2020 Computer Society of the Republic of China. All rights reserved.