A waterfall-style chart showing how each feature contributes to pushing a model prediction from a base value (expected model output) to the final predicted value. Horizontal bar segments extend right for positive SHAP values and left for negative SHAP values, stacking cumulatively so the viewer can trace the path from baseline to prediction. This is a core ML explainability visualization for explaining individual predictions, complementing the SHAP summary plot which shows feature effects across many samples.
#' anyplot.ai
#' shap-waterfall: SHAP Waterfall Plot for Feature Attribution
#' Library: ggplot2 3.5.1 | R 4.4.1
#' Quality: 95/100 | Created: 2026-09-09
library(ggplot2)
library(dplyr)
library(ragg)
library(scales)
set.seed(42)
# --- Theme tokens -------------------------------------------------------
THEME <- Sys.getenv("ANYPLOT_THEME", "light")
PAGE_BG <- if (THEME == "light") "#FAF8F1" else "#1A1A17"
INK <- if (THEME == "light") "#1A1A17" else "#F0EFE8"
INK_SOFT <- if (THEME == "light") "#4A4A44" else "#B8B7B0"
# Semantic exception (Imprint palette): SHAP explainability plots follow the
# domain convention of red = pushes prediction up, blue = pushes it down.
POSITIVE_COLOR <- "#AE3030" # Imprint position 5 — matte red
NEGATIVE_COLOR <- "#4467A3" # Imprint position 3 — blue
# --- Data -----------------------------------------------------------------
# Credit-scoring model: explaining one applicant's predicted default
# probability. base_value is the mean predicted probability across the
# training set; final_value is this applicant's actual prediction.
base_value <- 0.35
final_value <- 0.275
shap_df <- tibble::tibble(
feature = c(
"Credit Score", "Debt-to-Income Ratio", "Late Payments (12mo)",
"Employment Length", "Annual Income", "Loan Amount",
"Credit Utilization", "Age", "Open Credit Accounts", "Home Ownership"
),
shap_value = c(
-0.220, 0.150, 0.110, -0.090, -0.070,
0.060, 0.050, -0.030, -0.020, -0.015
)
)
waterfall_df <- shap_df %>%
arrange(desc(abs(shap_value))) %>%
mutate(
rank = row_number(),
y_pos = n() - rank + 1,
cum_end = base_value + cumsum(shap_value),
cum_start = cum_end - shap_value,
xmin_bar = pmin(cum_start, cum_end),
xmax_bar = pmax(cum_start, cum_end),
sign = factor(if_else(shap_value > 0, "positive", "negative"),
levels = c("positive", "negative")),
label = sprintf("%+.2f", shap_value),
label_x = if_else(shap_value > 0, xmax_bar + 0.006, xmin_bar - 0.006),
label_hjust = if_else(shap_value > 0, 0, 1)
)
connector_df <- waterfall_df %>%
arrange(rank) %>%
transmute(
x = cum_end,
y_top = y_pos - 0.35,
y_bottom = lead(y_pos) + 0.35
) %>%
filter(!is.na(y_bottom))
# --- Plot -------------------------------------------------------------------
p <- ggplot() +
geom_rect(
data = waterfall_df,
aes(xmin = xmin_bar, xmax = xmax_bar,
ymin = y_pos - 0.35, ymax = y_pos + 0.35, fill = sign),
color = PAGE_BG, linewidth = 0.4
) +
geom_segment(
data = connector_df,
aes(x = x, xend = x, y = y_top, yend = y_bottom),
color = INK_SOFT, linetype = "dotted", linewidth = 0.5
) +
geom_vline(xintercept = base_value, color = INK_SOFT, linetype = "dashed", linewidth = 0.6) +
geom_vline(xintercept = final_value, color = INK, linetype = "solid", linewidth = 0.8) +
geom_label(
data = waterfall_df,
aes(x = label_x, y = y_pos, label = label, hjust = label_hjust),
size = 3.2, color = INK, fill = PAGE_BG, label.size = 0,
label.padding = unit(0.12, "lines")
) +
annotate(
"text", x = base_value, y = 10.75,
label = paste0("Base value E[f(x)] = ", percent(base_value, accuracy = 0.1)),
hjust = 0.5, vjust = 0, color = INK_SOFT, size = 3.0
) +
annotate(
"text", x = final_value, y = 0.3,
label = paste0("Prediction f(x) = ", percent(final_value, accuracy = 0.1)),
hjust = 0.5, vjust = 1, color = INK, size = 3.2, fontface = "bold"
) +
scale_y_continuous(
breaks = waterfall_df$y_pos,
labels = waterfall_df$feature,
expand = expansion(add = c(0.8, 1.3))
) +
scale_x_continuous(
labels = percent_format(accuracy = 1),
expand = expansion(mult = c(0.08, 0.15))
) +
scale_fill_manual(
values = c(positive = POSITIVE_COLOR, negative = NEGATIVE_COLOR),
labels = c(positive = "Increases risk", negative = "Decreases risk"),
name = NULL
) +
labs(
title = "shap-waterfall · r · ggplot2 · anyplot.ai",
x = "Predicted Default Probability",
y = NULL
) +
theme_minimal(base_size = 8) +
theme(
plot.background = element_rect(fill = PAGE_BG, color = PAGE_BG),
panel.background = element_rect(fill = PAGE_BG, color = NA),
panel.grid.major.y = element_blank(),
panel.grid.minor = element_blank(),
panel.grid.major.x = element_line(color = INK, linewidth = 0.15),
axis.title.x = element_text(color = INK, size = 10),
axis.text.x = element_text(color = INK_SOFT, size = 8),
axis.text.y = element_text(color = INK, size = 9),
axis.ticks = element_blank(),
plot.title = element_text(color = INK, size = 12),
legend.position = "bottom",
legend.text = element_text(color = INK_SOFT, size = 8),
legend.key = element_rect(fill = PAGE_BG, color = NA)
)
# --- Save -------------------------------------------------------------------
ggsave(
filename = sprintf("plot-%s.png", THEME),
plot = p,
device = ragg::agg_png,
width = 8,
height = 4.5,
units = "in",
dpi = 400
)
Runnable source as JSON, for any HTTP client: https://api.anyplot.ai/specs/shap-waterfall/ggplot2/code. Any spec id and library id listed in llms-full.txt fit the same URL shape; every URL below is complete and callable.
{
"spec_id": "shap-waterfall",
"language": "r",
"library": "ggplot2",
"page": "https://anyplot.ai/shap-waterfall/r/ggplot2",
"hub": "https://anyplot.ai/shap-waterfall",
"code_json": "https://api.anyplot.ai/specs/shap-waterfall/ggplot2/code",
"spec_json": "https://api.anyplot.ai/specs/shap-waterfall",
"render_light_png": "https://storage.googleapis.com/anyplot-images/plots/shap-waterfall/r/ggplot2/plot-light.png",
"render_dark_png": "https://storage.googleapis.com/anyplot-images/plots/shap-waterfall/r/ggplot2/plot-dark.png",
"quality_score": 95.0,
"license": "MIT",
"guide": "https://anyplot.ai/llms.txt"
}Part of SHAP Waterfall Plot for Feature Attribution on anyplot.ai.