Jackson
Predicting Quality in Industrial Production Processes (303773)
*Adalbert Franz Xaver Wilhelm, Jacobs University Bremen
Keywords: model comparison, classification problem, imbalanced data, ensemble methods, LIME
This presentation summaries our experiences in developing predictive models for quality improvement in industrial production processes. Covering all stages from data preparation over data analysis to communicating the results we will focus on evaluative comparisons of different prediction techniques. We will mainly look at ensemble methods, support vector machines, neural networks and other machine learning tools. Besides presenting various ideas of comparing global measures, like accuracy and the area under the ROC curve, we also illustrate the use of the LIME – local interpretable model-agnostic explanation – approach. We will specifically focus on two aspects: how to best assess the specific requirements of highly imbalanced data and how to intertwine the knowledge and expectations of domain experts with the statistical characteristics of the classification approaches. Participants will get deeper insights into the functioning of machine learning, into approaches of evaluative comparisons between machine learning results, and the alignments of domain expert knowledge with data-generated information.