Interactive Concept Description with Bayesian Partition Models
Automotive companies are forced to extend and improve their product lineup continuously. However, increasing diversity, higher design complexity and shorter development cycles can produce new and unforeseen quality issues. In order to plan and implement efficient remedial actions, automotive engineers have to understand the root cause of these manufacturing or design related problems. The analysis of quality data provides valuable insights at all stages of this iterative process of detecting and resolving quality issues.
Based on the practical requirements, a new concept description approach, called interactive look-ahead decision trees, is developed. This approach combines interactive decision trees with Bayesian partition models and supports and interactive, iterative and intuitive way of causal investigation. The proposed method considers dependencies, in particular taxonomic and partonomic relationships among influencing variables and identifies the most likely, semantically meaningful partitions that are close to the concept that actually caused a quality issue. An evaluation on test data and real-world case studies illustrate how the approach can be used by engineers to investigate cause-effect relationships.