Modern vehicles, increasingly electrified and automated, have effectively become "computers on wheels," intensifying product complexity and competitive pressure. Concurrently, greater digitization offers opportunities to derive customer insights from large-scale vehicle data using Knowledge Discovery in Databases (KDD) and Data Mining (DM). Among these techniques, cluster analysis can reveal hidden subgroups that inform more customer-oriented product solutions. However, cluster analysis lacks a definitive ground truth, making it necessary to test numerous parameter settings, preprocessing steps, and clustering algorithms, and then interpret all plausible results.
The complexity of real-world customer data—such as heterogeneous, privacy-constrained vehicle usage signals—further complicates the selection of appropriate methodologies. Each combination of preprocessing and clustering steps must be analyzed to uncover patterns or groups, significantly increasing the time and manual effort required.
This paper presents a methodology to expedite the selection and evaluation of preprocessing and clustering configurations. By iteratively applying internal cluster validation indices on a representative data sample, it provides a quantitative basis for identifying promising approaches. Tested and validated on real-world vehicle usage data, this method streamlines the KDD and DM processes, reduces manual labor, and supports the development of robust, data-driven solutions.
Ultimately, this approach enables more efficient, customer-centric product development in the automotive domain, where understanding complex usage behaviors—from driving tasks to multimedia engagement—is critical for maintaining a competitive edge.