End-to-End Implementation of Automated Price Forecasting Applications

Forecasting prices of used construction equipment is challenging due to spatial and temporal price fluctuations. Automating this forecasting process using current market data is, therefore, highly desirable. A promising and common strategy is the application of machine learning (ML) techniques. However, small and medium-sized enterprise often struggle with the implementation of ML approaches due to a lack of ML expertise. In response, we demonstrate the potential of substituting manually created ML pipelines with automated machine learning (AutoML) solutions, which autonomously create the underlying pipelines. Therefore, we follow the CRISP-DM process to identify tasks requiring ML expertise. First, we dissect the ML pipeline into an machine learning and non-machine learning part and use AutoML to automate the former. Consecutively, we also automate the data preprocessing step, being part of the non-machine learning tasks, to further reduce the dependency on data processing expertise. Additionally, we implement a data-centric result evaluation, rating the reliability of the trained ML models. This approach supports the domain-driven creation of ML pipelines, democratizing the use of ML. To address all complex industrial requirements and showcase the practicality of our approach, we developed an innovative metric called method evaluation score. This metric encompasses key technical and non-technical parameters essential for domain experts to assess the quality and usability of the generated models. Based on this metric, we demonstrate in our case study that combining domain knowledge with AutoML and automatic preprocessing can reduce the reliance on ML experts for innovative small and medium-sized enterprise keen on adopting such technologies.