Sample-expand method for predicting the specified structure of microporous aluminophosphate

Abstract Imbalanced data sets often exist in many real-world fields and this problem has got more and more attention in recent years. In this paper, a sample-expand method is proposed as data pre-processing procedure to improve the predictive performance of the zeolite synthesis on imbalance data set. First, the data pre-processing is implemented for expanding samples by exploring the marginal structure of the given data set using k -nearest neighbor algorithm (KNN). Then, the expanded data set is input to support vector machines (SVM) for classification. Finally, Q times n -fold cross-validations procedure (CVs) is adopted to assess the prediction performance. The advantage of the data pre-processing is that it can obtain stable data set for establishing the training model and abide by the classification criteria of SVM, such that the improved predictive performance is achievable. Moreover, other classical machine learning methods are also presented to accomplish the prediction task. Compared experimental results demonstrate that SVM method can reach very satisfactory predictive accuracy on the pre-processing data set. Specially, the phase diagram of gel composition is provided as a guiding role for subsequent rational synthesis experiments.