RF-SVR-based prediction methodology for metal tube-bending rebound: Handling non-uniformity and limited sample challenges.

Abstract

This paper explores a prediction algorithm for determining the rebound angle of non-uniform and small-sample tubes. To address the issues of non-uniform and small-sample data, this paper proposes an algorithm based on Random Forest-Support Vector Regression (RF-SVR). Firstly, the polynomial feature generation method is introduced to solve the problem of non-uniform data. Secondly, after obtaining the data generated by the polynomial features, RF (Random Forest, an algorithm based on classification trees) is introduced to select the rebound features of the tubes, so that the features that have a profound influence on the rebound Angle can be retained. After obtaining the new data set, SVR (Support Vector Regression, an algorithm specifically designed for solving regression problems) is used to predict the rebound model of the bending of the metal tubes. The experimental results show that the RF-SVR method is superior to the traditional RF-BP and SVR methods, achieving higher prediction accuracy on small samples and non-uniform datasets.