A box plot (box-and-whisker plot) summarizes the distribution of a numeric variable — median, interquartile range, and outliers — as a single compact glyph, and critically, multiple box plots can be placed side by side to compare distributions across categorical groupings. This makes it well suited to the scenario described: comparing the spread and central tendency of performance scores across several models, further faceted by modality, in one readable figure. Box plots make skew, variance, and outlier prevalence immediately comparable across groups in a way a single summary statistic (like mean accuracy) cannot.
A histogram (B) shows the distribution of a single variable well but does not scale cleanly to side-by-side comparison across many model/modality combinations without becoming visually cluttered. A heatmap (A) is excellent for showing a matrix of values (e.g., mean score per model × modality pair) but represents point estimates, not distributions — it cannot convey variance or spread. A pie chart (D) is inappropriate for any continuous performance metric.
In practice, a violin plot — which overlays a kernel density estimate on the box plot's summary statistics — is often preferred when the underlying distribution's shape (e.g., bimodality) matters, but among the given options, the box plot is the correct choice for distributional comparison across groups.
[Reference: Data Analysis and Visualization domain — comparative distribution visualization, box plots vs. heatmaps., ]
