Prediction of steel fatigue strength based on mutual information feature selection and recursive feature elimination
YAO Lei
LI Zhen
WU Chuan
MENG Yafei
WANG Mengchao
Abstract：[Objective]To address the challenges in fatigue strength prediction caused by limited data and high-dimensional feature space,a method combining mutual information feature selection and recursive feature elimination(MIFS-RFE)is proposed.[Methods]Firstly,MIFS was used to identify crucial features for prediction.Subsequently,the remaining features were processed in the RFE stage.Through an iterative process,the most informative features were selected to ensure accurate fatigue strength prediction.The finalized feature subset was input into random forest regression(RFR),K-nearest neighbors regression(KNN),support vector regression(SVR),and multilayer perceptron(MLP)models for performance analysis.[Results]After optimization,the feature dimensionality was reduced from 25 to 13.During the test process,RFR,KNN,SVR,and MLP achieved R2 values of 0.977 7,0.972 5,0.961 3,and 0.976 6,respectively.Compared with the test results of all features,the proposed method increased the maximum R2 of the model by 0.020 8.Finally,based on the SHAP value,the influence of effective features was analyzed,and the effectiveness of the combination of MIFS and RFE was validated.The results indicate that the proposed method maintains high performance while reducing feature dimensionality,offering an optimized solution for fatigue strength prediction.
Keywords：Fatigue strengthMutual information feature selectionRecursive feature eliminationMachine learningOptimal feature subset
Publication Date：2026-01-31
Online Publishing Date：2026-09-12(First online date of this platform, not the publication date of the document)
Pages：7( 72-78 )