autovi R Package Automates Residual Plot Assessment with Computer Vision

Visual assessment of residual plots is a cornerstone of linear model diagnostics, but it's manual, inconsistent, and doesn't scale. The traditional lineup protocol—embedding the observed plot among null plots—reduces subjectivity but demands even more human effort. A new R package called autovi, developed by Weihao Li, Dianne Cook, and colleagues, tackles this by deploying a computer vision model that automatically evaluates residual plots. The model outputs a Visual Signal Strength (VSS) score, quantifying how much the residual pattern deviates from randomness. This supports analysts in assessing model fit without manual inspection. The approach builds on the lineup protocol's logic but automates the comparison, making it suitable for large-scale or repeated analysis.

autovi is accompanied by autovi.web, a Shiny application that lowers the barrier to use. The tool is designed to work with any linear model residuals, providing interpretable VSS scores and supporting information. According to the paper, published in the Australian & New Zealand Journal of Statistics (Vol. 68, Issue 1, 2026), the automation addresses key pain points: scalability and analyst bias. By replacing subjective human judgment with a consistent algorithmic evaluation, it enables reproducible diagnostics across teams. The code and app are openly available via arXiv (arXiv:2606.24236). For data scientists and statisticians, autovi represents a practical step toward integrating computer vision into routine statistical workflows, particularly in high-volume environments like automated machine learning pipelines.