Automated Machine Learning for Steel Production: A Case Study of TPOT for Material Mechanical Property Prediction

Abstract

Steel is a highly important product for which there has been an ever increasing demand. A critical challenge for steel production is the development of accurate methods for material mechanical property prediction (MMPP). The increasing availability of data has enabled the use of machine learning (ML) to address this challenge. Recently various ML- based methods have been developed, amongst which, however, no single one can achieve competitive results in all product prediction problems. In this context, automated machine learning (AutoML) has been proposed to automatically solve the model selection problem in a way that is more efficient than traditional machine learning. This method is particularly useful for situations in which a considerable amount of human expert knowledge is required to construct specialized ML models. Although AutoML methods are useful and effective, their applications are still limited and difficult due to the lack of analysis of their strengths and weaknesses. Thus, this paper aims to analyze the application of AutoML methods to MMPP processes with datasets collected in the real world by comparing two state-of-the-art AutoML methods (i.e. Auto-Sklearn and TPOT). Results show that AutoML can be considered as a powerful approach to address the model selection problem in MMPP. In particular, TPOT demonstrates the advantages of computational consumption and competitive accuracy.