Evolutionary design of a visual analytics interface to study predictive patterns in high dimensional data

Organizations increasingly rely on data mining to discover predictive patterns that inform decision-making. However, the vast number of patterns produced by mining algorithms, especially in high-dimensional datasets, can overwhelm analysts. To address this challenge, we present the evolutionary design of a visual analytics interface that enables users to explore and interpret mined predictive patterns effectively. Starting from a projection-based prototype rooted in research tools, we iteratively refined the interface through feedback from data analysts and real-world usage. Each design iteration – projection maps, bubble plots, and finally a card-based layout – was shaped by insights into user comprehension, trust, and pattern comparison needs. The final interface supports overview-to-detail exploration, pattern organization, and contextual analysis, as demonstrated in a case study on student attrition in Computer Science. We reflect on the lessons learned across iterations and offer guidance for designing interpretable pattern exploration tools for high-dimensional data. • A comparative analysis of three visual interface designs for pattern exploration in high-dimensional data. • Design insights on how visualization techniques influence pattern interpretability and user trust. • A production-ready interface demonstrated through a real-world case study. • Design guidance for building effective pattern exploration tools that bridge machine learning and human sensemaking.