A multi-objective optimizer-based model for predicting composite material properties

Abstract Composite material property testing usually requires multiple experiments if multiple variables, is time-consuming. The practice in many fields indicates that machine learning models have great potential in solving this problem because they can predict the material properties through the existing data. In this work, a hybrid model combining multi-objective grey wolf optimizer and support vector machine is proposed to predict composite material properties in six datasets. Among them, three datasets have time series characteristics, and the rest do not. The results reveal that the proposed model performs well in the property prediction of composite materials, the mean absolute percentage error of the predictions ranged from 0.14% to 5.574%. Discussions indicate that the more data in the training set, the better the model’s prediction performance. Moreover, machine learning models have great potential in composite material property testing and new material design. The correlation between different variables and material properties is analyzed, and the influence of input variables on prediction is also discussed. The results imply that if factors with weak linear correlation are not used as input, the model’s prediction accuracy in some datasets may be improved.