Visualization Intelligence Toolkit

Provides automated data profiling and visualization recommendation tools for exploratory data analysis (EDA). The package inspects a data frame, classifies each variable as numeric, categorical, or datetime, and computes distributional statistics including skewness, excess kurtosis, Shapiro-Wilk normality test p-values (Royston 1995) , and cardinality. Based on these profiles, the package scores and ranks candidate chart types (histogram, density, boxplot, violin, bar, lollipop, scatter, hexbin, line, heatmap) using an internal taxonomy that accounts for variable types, analysis goal, skewness, cardinality, and data size. Additional functions audit existing ggplot objects for common visualization integrity problems such as truncated axes, lie factors, overplotting, and dual y-axes following principles described in Tufte (1983, ISBN:0961392142) and Wilke (2019) < https://clauswilke.com/dataviz/>. The package also checks visualizations for accessibility issues and suggests alternative chart types for a given plot.

vizintel

vizintel is an R package built as a course project. It is a toolkit for understanding data and checking whether a plot communicates it correctly.

Five core visualization intelligence functionalities are available:

1. viz_recommend() -- suggests chart types based on column profiles
2. viz_audit() -- takes a finished ggplot and checks it for common mistakes
3. viz_explore() -- explores relationships between numeric variables using correlation analysis and a heatmap
4. viz_alternatives() -- suggests alternative visualization types for an existing ggplot
5. viz_accessibility() -- checks a finished ggplot for common accessibility issues

The package also contains internal helper functionality: profile_data() for profiling data frames.

Installation

# from the local repo
devtools::install("path/to/R-package")

Requires R >= 4.0. Packages e1071, ggplot2, dplyr, and gridExtra are installed automatically as dependencies. The hexbin package is suggested (needed only for hexbin chart panels in plot.viz_recommend_result()).

What it does

1. Getting chart recommendations

viz_recommend() looks at your data frame's columns, works out what type each one is, and suggests the most appropriate chart for each.

library(vizintel)

# Recommendations for every column
viz_recommend(mtcars)

# Recommendations for a specific pair of columns
viz_recommend(mtcars, x = "wt", y = "mpg")

# Force a particular goal
viz_recommend(mtcars["mpg"], goal = "distribution")

Sample output:

== vizintel chart recommendations ==

  mpg                  ->  histogram      (score: 1.00)
                           "Shows the full distribution of a numeric variable."
                       ->  density        (score: 0.95)
                           "Smoother alternative to a histogram for continuous data."
                       ->  log_histogram  (score: 0.90)
                           "Better than a histogram when the data is heavily right- or left-skewed."

The goal parameter accepts: "distribution", "compare", "trend", "relationship", "composition". When left NULL, the goal is inferred from the column type.

Plotting recommendations

viz_recommend() stores the data inside its result, so you can call plot() directly on it — no need to pass the data again. Each recommended chart type is rendered as a real ggplot panel and all panels are arranged in a single view using gridExtra::grid.arrange():

res <- viz_recommend(mtcars["mpg"])
plot(res)                          # grid of histogram / density / log_histogram

res2 <- viz_recommend(mtcars, x = "wt", y = "mpg")
plot(res2)                         # scatter + hexbin panels side-by-side

plot(res, ncol = 1)               # pass extra args to grid.arrange()

Each panel title shows the chart type and its fit score, e.g. histogram (score: 1.00).

2. Auditing a plot

viz_audit() takes any finished ggplot and runs four checks on it. Every issue now includes a plain-English explanation field, printed under the technical message as "In plain terms: ..." so non-expert readers can understand what the problem means.

library(ggplot2)

# A bar chart where the y-axis does not start at 0
p <- ggplot(data.frame(x = c("A", "B"), y = c(50, 80)), aes(x, y)) +
  geom_col() +
  coord_cartesian(ylim = c(40, 85))

report <- viz_audit(p)
print(report)

Output:

== vizintel audit report ==
Geoms detected: GeomCol
Issues: 1 critical, 0 warning, 0 info

x [CRITICAL] Y-axis starts at 40, not 0 — bar heights are visually exaggerated.
   In plain terms: Because the Y-axis does not start at zero, the bars appear
   much taller relative to each other than the actual values justify. A small
   real difference can look enormous, making the chart misleading at a glance.

summary(report)
# [FAIL] 1 issue(s) -- 1 critical / 0 warning / 0 info

Each issue object also carries the explanation as a field you can access programmatically:

report$issues[[1]]$explanation

The four checks it runs:

3. Profiling a data frame

profile_data() is internal -- used by viz_recommend() and viz_explore() to identify and summarise columns. You can call it directly with ::: if you want to inspect the raw column statistics.

df <- data.frame(
  score = c(72, 85, 61, 90, 78),
  grade = c("B", "A", "C", "A", "B"),
  date  = seq(as.Date("2024-01-01"), by = "week", length.out = 5)
)

vizintel:::profile_data(df)

Returns one row per column with: type classification, missing count, skewness, kurtosis, normality test result, cardinality (categorical), regularity (datetime).

4. Exploring relationships between variables

viz_explore() explores relationships between numeric variables using a correlation matrix and a correlation heatmap.

library(vizintel)

result <- viz_explore(mtcars)

# View the correlation matrix
result$correlation

# Display the correlation heatmap
result$plot

The function identifies numeric variables automatically and calculates the correlation between each pair of numeric variables.

The result contains:

At least two numeric variables are required. Non-data-frame input and data with fewer than two numeric variables raise clear errors.

5. Finding visualization alternatives

viz_alternatives() suggests alternative visualization types for an existing ggplot based on its geometry and statistical transformation.

library(ggplot2)

p <- ggplot(mtcars, aes(wt, mpg)) +
  geom_point()

viz_alternatives(p)

Example result:

[1] "2D Density Plot"
[2] "2D Histogram"

The alternatives depend on the current visualization:

Histograms are distinguished from regular bar plots using the plot's statistical transformation.

6. Checking visualization accessibility

viz_accessibility() checks an existing ggplot for common accessibility issues that may make a visualization difficult to interpret.

library(ggplot2)

p <- ggplot(mtcars, aes(wt, mpg, colour = factor(cyl))) +
  geom_point()

viz_accessibility(p)

The function checks for:

The function returns a list containing an issues element. Each issue contains a severity level, check name, and descriptive message.

If no issues are detected, the function returns:

No obvious accessibility issues detected.

The function provides practical checks for common accessibility concerns. It does not perform full WCAG compliance testing.

Future work

The five core visualization intelligence functionalities are now implemented:

• viz_recommend() -- visualization recommendation
• viz_audit() -- visualization integrity auditing
• viz_explore() -- multivariate exploration
• viz_alternatives() -- visualization alternatives
• viz_accessibility() -- visualization accessibility

Future work may focus on improving the existing checks, expanding supported visualization types, and adding more test coverage.

Files in the package

R/
  profile-data.R           profile_data() and classify_column() -- internal engine
  extract-plot-geometry.R  shared ggplot geometry extraction for 
                           viz_audit(), viz_alternatives(), and
                           viz_accessibility()
  viz-recommend.R          viz_recommend(), plot.viz_recommend_result(),
                           taxonomy table, scoring logic, and print method
  viz-audit.R              viz_audit(), four check functions (each with an
                           explanation field), print/summary methods
  viz-explore.R            viz_explore(), correlation analysis and heatmap
  viz-alternatives.R       viz_alternatives(), visualization alternatives
  viz-accessibility.R      viz_accessibility(), accessibility checks

tests/testthat/
  test-profile-data.R      7 tests for profile_data()
  test-viz-recommend.R     12 tests for viz_recommend() and plot()
  test-viz-audit.R         11 tests for viz_audit() (incl. explanation field)
  test-viz-explore.R       5 tests for viz_explore()
  test-viz-alternatives.R  5 tests for viz_alternatives()
  test-viz-accessibility.R 5 tests for viz_accessibility()

Authors

T Pooja and Ronesha Navaneethan — MIT License 2026

Supervisor: Dr Amrit Pal