Benchmarking AutoML-Supported Lead Time Prediction
Janek Bender, Martin Trat, Jivka Ovtcharova
Abstract
In manufacturing companies, especially in small and medium enterprises (SME) in the make-to-order domain, predicting accurate lead times for highly customisable products poses a challenge. One avenue to tackle this challenge is using machine learning to create data-driven models. With the advent of Automated Machine Learning (AutoML), there is a potential also for non-data-scientists in the SMEs to utilise machine learning in a self-sufficient manner. In this paper, we benchmarked three AutoML solutions on data from two SMEs and compared their models with manual predictions currently in productive use as well as simple mean models. We furthermore investigated how these solutions support the model development process. While we found that the models generated by the AutoML solutions outperformed manual predictions, labour-intensive parts of the model development process such as data understanding and pre-processing as well as feature engineering were not supported sufficiently by the benchmarked solutions.