Prediction for Rational Synthesis Based on Weighted Feature Selection Method

In this paper, a novel integrated feature selection model is proposed to analyze the relationship between the synthesis factors and the specific resulting structure on the database of AlPO synthesis. Concretely, the proposed model can select the most significant synthesis factors affecting the formation of a (6,12)-ring-containing structure by combining multiple feature selection methods. Firstly, eight feature selection methods are employed to prerank the synthesis factors based on the predictive performance of support vector machine. Then, a weighted fusion mechanism is presented to rerank the results. Finally, sequential forward floating search method is utilized to select the most significant synthesis factors in view of the highest predictive performance. A large number of experimental results show that the proposed model is efficient and feasible. The predictive accuracy can reach 86.47 % with the selected 10 factors among 21 synthesis factors. The selection and ranking results also give a rational understanding for AlPO synthesis. More specifically, a proportional relationship among gel composition parameters is recommended based on the result of feature selection, which has important guiding significance for the rational design and synthesis.