Aiming at the disadvantages of rationality and adaptability of linear dimensionless method, as well as the complexity of constructing polyline and curve dimensionless method, this paper proposes an Interval-based Non-dimensionalization Method (IBNM). IBNM can be assembled in to polyline IBNM or curve IBNM based on the critical points formed by interval partition. Interval division can be classified based on the existing index data, which is scientific, reasonable, simple and practical. Taking example of the prediction of PM2.5 air quality grade in Guangzhou, this paper constructs four models to predict air quality grade, such as, Support Vector Regression (SVR), Random Forest, Extremely Random Tree, and Gradient Lifting Regression Tree model. Meanwhile, the data is processed with extremum method, polyline IBNM and curve IBNM. The results show that the accuracy of polyline IBNM and curve IBNM is better than extremum method.