A novel approach to multi-attribute predictive analysis based on rough fuzzy sets

Predictive analysis is vital for decision management especially when involving in multiple attributes information systems. The correlation between attributes reflects that there exists a certain association between attributes and objects. How to effectively use the correlation between attributes to guide the prediction is a hot research topic in predictive analysis. In this paper, based on the designed attribute-oriented rough fuzzy set (RFS) model, a new multi-attribute predictive analysis approach is put forward for dealing with multi-attribute fuzzy information systems. This approach uses delta-fuzzy similarity classes to gather attributes with strong correlation to construct a new RFS model. Moreover, two prediction directions are constructed on the basis of the lower and upper approximations of the new RFS model. By analyzing the cosine distances between alternative and pessimistic and optimistic prediction directions, a trend predictive function is designed to forecast the development trend of the alternative. In the process of model building, trend prediction is carried out based on certain semantic information, which shows that the prediction model has certain interpretability. To evaluate the performance of the proposed multi-attribute predictive analysis model, experiments with UCI datasets are conducted for analysis and discussion. The obtained experimental results indicate that the established predictive analysis model is feasible and satisfactory.