Streamlining the development of data-driven industrial applications by automated machine learning
Dominik Kißkalt, Andreas Mayr, Benjamin Lutz, Annelie Rögele, Jörg Franke
Abstract
Machine learning has often proven superior to traditional white-box modeling in industrial application scenarios. Yet the determinism in finding a solution close to the theoretical optimum is low due to human factors in the development process. Automated machine learning (AutoML), on the other hand, allows a complete automation of the machine learning pipeline from feature extraction and preprocessing to model selection and hyperparameter optimization. Using a popular open dataset, this paper exemplifies how AutoML can streamline the development of data-driven industrial applications. As a benchmark, results from existing approaches on the same dataset are used.
Topics & Concepts
Machine learningArtificial intelligenceComputer sciencePipeline (software)AutomationProcess (computing)Data pre-processingPreprocessorBenchmark (surveying)HyperparameterEngineeringMechanical engineeringProgramming languageGeographyGeodesyOperating systemIndustrial Vision Systems and Defect DetectionMachine Learning and Data ClassificationFault Detection and Control Systems