A Hybrid Rule Mining and Machine Learning Approach for BI Dashboard Design

Abstract

Modern business intelligence (BI) platforms demand sophisticated yet intuitive dashboard designs that adapt to diverse user roles, dataset schemas, and analytical objectives. Manual dashboard authoring is time-consuming, requires domain expertise, and often produces inconsistent visual encodings. This paper presents DashCraft BI — a Dashboard Design Mining and Recommendation System — implemented as a fully interactive web application built on Streamlit, Plotly, and Python. DashCraft mines six evidence-based visualization design rules from uploaded CSV datasets, infers column data types automatically, adapts chart selections to user roles (Executive vs. Analyst), and auto-generates role-aware dashboards. The system further provides a live streaming analytics module with moving-average overlays and threshold-based alerting. Evaluation on built-in benchmark datasets demonstrates high rule-confidence scores (79%–96%) and end-to-end latency well within interactive thresholds. DashCraft bridges the gap between raw tabular data and publication-quality dashboards without requiring manual chart specification.