Using meta-learning for automated algorithms selection and configuration: an experimental framework for industrial big data

Abstract

AbstractAdvanced analytics are fundamental to transform large manufacturing data into resourceful knowledge for various purposes. In its very nature, such “industrial big data” can relay its usefulness to reach further utilitarian applications. In this context, Machine Learning (ML) is among the major predictive modeling approaches that can enable manufacturing researchers and practitioners to improve the product quality and achieve resource efficiency by exploiting large amounts of data (which is collected during manufacturing process). However, disposing ML algorithms is a challenging task for manufacturing industrial actors due to the prior specification of one or more algorithms hyperparameters (HPs) and their values. Moreover, manufacturing industrial actors often lack the technical expertise to apply advanced analytics. Consequently, it necessitates frequent consultations with data scientists; but such collaborations tends to cost the delays, which can generate the risks such as human-resource bottlenecks. As the complexity of these tasks increases, so does the demand for support solutions. In response, the field of automated ML (AutoML) is a data mining-based formalism that aims to reduce human effort and speedup the development cycle through automation. In this regard, existing approaches include evolutionary algorithms, Bayesian optimization, and reinforcement learning. These approaches mainly focus on providing the user assistance by automating the partial or entire data analysis process, but they provide very limited details concerning their impact on the analysis. The major goal of these conventional approaches has been generally focused on the performance factors, while the other important and even crucial aspects such as computational complexity are rather omitted. Therefore, in this paper, we present a novel meta-learning based approach to automate ML predictive models built over the industrial big data. The approach is leveraged with development of, AMLBID, an Automated ML tool for Big Industrial Data analyses. It attempts to support the manufacturing engineers and researchers who presumably have meager skills to carry out the advanced analytics. The empirical results show that AMLBID surpasses the state-of-the-art approaches and could retrieve the usefulness of large manufacturing data to prosper the research in manufacturing domain and improve the use of predictive models instead of precluding their outcomes.