Concrete Strength Drivers Dashboard

SHAP-based ranking and non-linear dependence patterns for concrete compressive strength

Dominant Driver
Age
Mean |SHAP| importance = 7.77
Highest overall model influence
Second Strongest
Water/Binder Ratio
Mean |SHAP| importance = 6.39
Strong inverse strength signal
Cement Taper Point
~355 kg/m³
LOWESS slope starts to soften materially
More cement helps, then gains thin out
Near-Peak Age Effect
~90 days
Smoothed age effect reaches ~90% of peak
Late curing still matters most
Executive Readout

The SHAP ranking says the model is driven primarily by curing age and water/binder ratio, with cement as the third most influential lever. Together, the top two features account for 62.4% of the total importance, and the top three account for 77.5%.

The dependence plots show three distinct shapes: a steadily rising age effect, a cement response that eventually tapers, and a sharply negative response as water/binder ratio moves higher.

1,030 observations per dependence plot Age turns clearly positive at ~56 days Water/binder turns net negative above ~0.44
How to Read the Dependence Plots

Each point is one observation from the SHAP analysis. Values above zero indicate a feature level that pushes predicted compressive strength upward relative to the model baseline; values below zero pull it downward.

  • The bold trendline is a LOWESS smoother, used here to expose non-linear structure without assuming a straight-line effect.
  • For cement, the curve climbs strongly at lower dosages and then flattens after the mid-300 kg/m³ range.
  • For age, the model penalizes very early curing, then increasingly rewards longer curing windows, with most of the upside in place by roughly 90 days.
  • For water/binder ratio, lower ratios are strongly beneficial, while higher ratios quickly become a drag on predicted strength.
SHAP Feature Importance Ranking
Mean absolute SHAP values show which mix and curing variables matter most to the model.
Cement Dependence Plot
The smoothed curve highlights diminishing marginal lift as cement dosage rises.
Age Dependence Plot
Very early ages are penalized, while longer curing strongly improves the model’s predicted strength contribution.
Water/Binder Ratio Dependence Plot
A low water/binder ratio is favorable; beyond the threshold, the contribution turns negative.
Feature Ranking Detail
Rank Feature Mean |SHAP| Share Relative
Relative compares each feature’s mean |SHAP| against the top-ranked feature.
Practical Interpretation

The dashboard separates importance from shape. Importance tells you which features the model relies on most overall; the dependence curves reveal where each feature helps, hurts, or starts to plateau.

  • Age is the clearest overall strength driver in this model.
  • Water/binder ratio is nearly as important and has the sharpest negative slope once the ratio rises past the favorable zone.
  • Cement still contributes positively at high dosages, but the lowess curve suggests that the marginal return becomes much smaller after roughly 355 kg/m³.
  • These SHAP patterns reflect the fitted model’s behavior across the observed data distribution and are best used to understand trade-offs, thresholds, and non-linear responses.