基于RFE-SHAP的具有可解释性纱线质量预测研究OA

To optimize the process of feature selection for yarn quality prediction,further eliminate the redundant features in the conditions of small samples,improve the accuracy and reliability of the subsequent prediction process,an interpretable method of yarn quality prediction based on recursive feature elimination algorithm(RFE)and SHAP was proposed,namely RFE-SHAP.Firstly,RFE was selected as the iterative method of feature selection,and support vector regression(SVR)was used as the evaluator.Then,SHAP technology was introduced to quantify the marginal contribution of the original features to the yarn strength and hairiness H value to assist in feature selection and provide a more intuitive and explanatory strategy for feature selection.Finally,a neural network was combined to construct a prediction model for yarn strength and hairiness H value.The experimental results showed when the subset of optimal features selected by RFE-SHAP was used as the input of the prediction model of yarn strength and hairiness H value,the effect of the modle multiple evaluation indices was improved,and the average absolute percentage error of the two yarn quality indices prediction was not exceed 3%.It is considered tha the method have higher feasibility and can improve the prediction performance of the model to a certain extent.