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: makie 0.21.9 | Julia 1.11.9
# Quality: 86/100 | Created: 2026-09-09
using CairoMakie
using Colors
using Printf
# --- Theme tokens -----------------------------------------------------------
THEME = get(ENV, "ANYPLOT_THEME", "light")
PAGE_BG = THEME == "light" ? colorant"#FAF8F1" : colorant"#1A1A17"
INK = THEME == "light" ? colorant"#1A1A17" : colorant"#F0EFE8"
INK_SOFT = THEME == "light" ? colorant"#4A4A44" : colorant"#B8B7B0"
# Imprint palette semantic anchors: matte red for positive push, blue for negative
POSITIVE_COLOR = colorant"#AE3030"
NEGATIVE_COLOR = colorant"#4467A3"
# --- Data ---------------------------------------------------------------
# Credit scoring model explaining a single loan applicant's predicted default
# probability. Features ranked by descending absolute SHAP magnitude.
features = [
"Debt-to-Income Ratio",
"Credit Utilization",
"Recent Missed Payments",
"Income Stability",
"Credit History Length",
"Age",
"Number of Open Accounts",
"Employment Length",
"Loan-to-Value Ratio",
"Number of Credit Inquiries",
]
shap_values = [7.2, 5.1, 4.3, -3.8, -3.1, -2.4, 1.8, -1.5, 1.2, 0.9]
base_value = 18.4
n = length(features)
ends = base_value .+ cumsum(shap_values)
starts = ends .- shap_values
final_value = ends[end]
# Position n (top) holds the largest-magnitude feature; position 1 (bottom)
# holds the smallest, matching the descending-magnitude ranking above.
positions = collect(n:-1:1)
bar_colors = [v >= 0 ? POSITIVE_COLOR : NEGATIVE_COLOR for v in shap_values]
# --- Plot -----------------------------------------------------------------
fig = Figure(
resolution = (1600, 900),
fontsize = 14,
backgroundcolor = PAGE_BG,
)
lo = minimum(vcat(starts, ends, [base_value])) - 5.0
hi = maximum(vcat(starts, ends, [final_value])) + 5.0
ax = Axis(
fig[1, 1];
title = "shap-waterfall · julia · makie · anyplot.ai",
titlesize = 20,
titlecolor = INK,
xlabel = "Predicted Default Probability (%)",
xlabelsize = 14,
xlabelcolor = INK,
xticklabelsize = 12,
xticklabelcolor = INK_SOFT,
yticks = (1:n, reverse(features)),
yticklabelsize = 13,
yticklabelcolor = INK_SOFT,
ylabel = "Feature",
ylabelsize = 14,
ylabelcolor = INK,
backgroundcolor = PAGE_BG,
topspinevisible = false,
rightspinevisible = false,
leftspinevisible = false,
leftspinecolor = INK_SOFT,
bottomspinecolor = INK_SOFT,
xgridvisible = true,
ygridvisible = false,
xgridcolor = RGBAf(INK.r, INK.g, INK.b, 0.15),
xminorgridvisible = false,
yminorgridvisible = false,
)
xlims!(ax, lo, hi)
ylims!(ax, 0.2, n + 1.3)
# Dotted connector lines between consecutive cumulative segments
for i in 1:(n - 1)
lines!(
ax,
[ends[i], ends[i]],
[positions[i], positions[i + 1]];
color = INK_SOFT,
linestyle = :dot,
linewidth = 1.5,
)
end
# Base value and final prediction reference lines, stopped below the label
# row (as plain lines! with an explicit y-extent) so the dashed lines never
# cross through their own annotation text above.
line_top = n + 0.4
lines!(ax, [base_value, base_value], [0.2, line_top]; color = INK_SOFT, linestyle = :dash, linewidth = 1.5)
lines!(ax, [final_value, final_value], [0.2, line_top]; color = INK, linestyle = :dash, linewidth = 1.5)
scatter!(ax, [base_value], [line_top]; marker = :utriangle, markersize = 10, color = INK_SOFT)
scatter!(ax, [final_value], [line_top]; marker = :utriangle, markersize = 10, color = INK)
text!(
ax, base_value, n + 0.9;
text = @sprintf("Base value: %.1f%%", base_value),
color = INK_SOFT, fontsize = 13, align = (:center, :bottom),
)
text!(
ax, final_value, n + 0.5;
text = @sprintf("Prediction: %.1f%%", final_value),
color = INK, fontsize = 13, align = (:center, :bottom), font = :bold,
)
# Waterfall bars (horizontal, floating from cumulative start to cumulative end)
barplot!(
ax,
positions,
ends;
fillto = starts,
direction = :x,
color = bar_colors,
width = 0.62,
strokewidth = 0,
)
# Numeric SHAP value labels beside each bar
for i in 1:n
label = @sprintf("%+.1f", shap_values[i])
x_offset = shap_values[i] >= 0 ? ends[i] + 0.5 : ends[i] - 0.5
text_align = shap_values[i] >= 0 ? (:left, :center) : (:right, :center)
text!(
ax, x_offset, positions[i];
text = label, color = INK, fontsize = 13, align = text_align,
)
end
# Legend explaining bar color semantics
legend_elems = [PolyElement(color = POSITIVE_COLOR), PolyElement(color = NEGATIVE_COLOR)]
legend_labels = ["Increases prediction", "Decreases prediction"]
Legend(
fig[2, 1], legend_elems, legend_labels;
orientation = :horizontal, framevisible = false,
labelcolor = INK, labelsize = 13, backgroundcolor = PAGE_BG,
)
rowsize!(fig.layout, 1, Relative(0.92))
# --- Save -------------------------------------------------------------------
save("plot-$(THEME).png", fig; px_per_unit = 2)
Runnable source as JSON, for any HTTP client: https://api.anyplot.ai/specs/shap-waterfall/makie/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": "julia",
"library": "makie",
"page": "https://anyplot.ai/shap-waterfall/julia/makie",
"hub": "https://anyplot.ai/shap-waterfall",
"code_json": "https://api.anyplot.ai/specs/shap-waterfall/makie/code",
"spec_json": "https://api.anyplot.ai/specs/shap-waterfall",
"render_light_png": "https://storage.googleapis.com/anyplot-images/plots/shap-waterfall/julia/makie/plot-light.png",
"render_dark_png": "https://storage.googleapis.com/anyplot-images/plots/shap-waterfall/julia/makie/plot-dark.png",
"quality_score": 86.0,
"license": "MIT",
"guide": "https://anyplot.ai/llms.txt"
}Part of SHAP Waterfall Plot for Feature Attribution on anyplot.ai.