A novel sequential three-way decisions model based on penalty function

Abstract In the sequential three-way decisions (S3WD) model, classification accuracy is an important issue. Based on the idea of minimum misclassification, loss functions have been used to calculate the thresholds. A lot of achievements have been achieved in the related research by using the idea of minimum misclassification. However, few researchers have focused on minimizing the misclassification by improving the classification precision in different granularity layers. In this paper, from the classification precision difference between two adjacent granularity layers, a new sequential three-way decisions model based on penalty function (S3WDPF) is proposed to improve the classification accuracy by modifying cost parameters. First, two types of negative benefit classification in the S3WD model are defined. Next, based on the idea of optimization, the penalty function is devised to optimize the cost parameters. Then, from the viewpoint of Bayesian minimum risk decision, a decision rule of mutually exclusive decision thresholds is designed, and the rules of three-way decisions are deduced. Further, the change rules of decision thresholds after modifying cost parameters are discussed in detail. Finally, the experimental results show that the performance of the S3WDPF model has improved classification accuracy compared with the current existing models.