Towards a data science toolbox for industrial analytics applications

Abstract Manufacturing companies today have access to a vast number of data sources providing gigantic amounts of process and status data. Consequently, the need for analytical information systems is ever-growing to guide corporate decision-making. However, decision-makers in production environments are still very much focused on static, explanatory modeling provided by business intelligence suites instead of embracing the opportunities offered by predictive analytics. We develop a data science toolbox for manufacturing prediction tasks to bridge the gap between machine learning research and concrete practical needs. We provide guidelines and best practices for modeling, feature engineering and interpretation. To this end, we leverage tools from business information systems as well as machine learning. We illustrate the usage of this toolbox by means of a real-world manufacturing defect prediction case study. Thereby, we seek to enhance the understanding of predictive modeling. In particular, we want to emphasize that simply dumping data into “smart” algorithms is not the silver bullet. Instead, constant refinement and consolidation are required to improve the predictive power of a business analytics solution.