Swarm intelligence-based optimization of machine learning models for diverse industrial problems: a comparative study

Abstract

Abstract With the rapid development of artificial intelligence, machine learning has emerged as a promising approach to tackle complex problems in various industries. It is particularly evident in situations where limited data samples are available, such as when developing new materials, or in high-speed machining scenarios where computational efficiency and result reliability are equally crucial. In this work, we have selected five widely recognized machine learning models (e-SVR, v-SVR, BP, Random Forest, Extreme Learning Machines) as baseline methods. We employ Snake Optimizer, a recently popular swarm intelligence algorithm, to optimize the parameters of each model individually. The Mean Squared Error is utilized as an evaluation fitness measure to adaptively optimize and determine the optimal hyperparameter combination for each method. Consequently, five hybrid models are established. To thoroughly evaluate the performance and applicability of these models, predictive experiments are conducted using temperature data of permanent magnet synchronous motors, energy consumption data from steel companies and real-estate price data of Chicago, representing three distinct scenarios. The learning ability and generalization capability of these models are tested and assessed using four error measures: Mean Absolute Error, R-Squared, Root Mean Squared Error, and Computation Time. Moreover, comparison and discussion between the hybrid models and their respective basic models are also conducted, analysing the effectiveness of SO algorithm in optimization. Through analysis and discussion, a comprehensive understanding of the characteristics and applicability of them is provided. Our objective is to provide scientific researchers and engineers with valuable insights into different methods and their characteristics.