ImputeViz Dashboard Diagnoses Missing Data and Compares 5 Imputation Methods

Missing data is a pervasive problem in research, often biasing results and forcing analysts to make difficult choices about imputation. ImputeViz, presented in a new arXiv paper from Stony Brook University, addresses this with an integrated visual analytics dashboard that makes the entire process transparent. The tool combines five imputation methods — MICE, Random Forest, XGBoost, kNN, and a new geographically informed kNN variant called gKNN — within a single interactive environment.

A key innovation is gKNN, which blends socioeconomic and spatial distances when estimating missing values, and exposes the donor contributions so analysts can see which geographic regions drive each estimate. The dashboard’s coordinated views include heatmaps of missingness patterns, co-missingness summaries, and distributional diagnostics to help classify missingness mechanisms (MCAR, MAR, MNAR). A Method Comparison Summary reports MAE, RMSE, Delta RMSE, and runtime for each algorithm on the current target and mask, along with variable-level discrepancy views. Cached results and locked axis scales reduce cognitive load when switching methods, making cross-method comparison a first-class visual task.

Case studies demonstrate how ImputeViz helps researchers surface sensitive variables, compare imputation strategies, and assess model robustness. The system is particularly useful for social science and public health datasets where geographic context matters. By providing provenance-based visual accountability and allowing users to drill into where methods disagree, ImputeViz aims to turn a traditionally opaque step in data analysis into an informed, collaborative decision process.