A Heat Transfer Search Algorithm Optimized Support Vector Regressor for Predicting Transportation Cost in the Steel Industry

Abstract

Different machine learning (ML) models are used in various engineering applications for prediction purposes. The performance of ML models is often governed by their hyperparameters. Support vector regression (SVR) is a widely accepted ML model for predicting transportation costs across various engineering fields, as it can learn non-linear relationships with the help of a kernel, even with small to medium-sized datasets. Therefore, the present work uses SVR to model transportation cost in the steel industry, with features that account for sustainability. However, SVR also includes hyperparameters for the kernel function and the regularization coefficient. The present study uses a parameterless heat transfer search (HTS) algorithm to optimize the hyperparameter of the SVR. The results show that the HTS algorithm not only provides better metrics on test data but also makes the model generalizable for practical implementation.