Simple Binning Algorithm and SimDec Visualization for Comprehensive Sensitivity Analysis of Complex Computational Models

This paper formalizes and stress-tests the Simple Binning Approach (SBA) for computing variance-based sensitivity indices behind SimDec.

Most global sensitivity analysis studies opt for Saltelli’s implementation of Sobol’ indices. SBA follows the original intuition behind variance-based sensitivity:
bin the input, compute conditional output means, take their variance, normalize.

The paper demonstrates that SBA:

• Requires substantially fewer model evaluations than classic Sobol’ estimators
• Captures second-order effects directly
• Works with dependent inputs without transformation
• Preserves the conservation property (sum of indices ≈ 1)
• Operates on a single simulation dataset
• Can be applied to empirical data

Testing includes:

• The portfolio benchmark model (comparison against Saltelli estimators)
• The Ishigami function (nonlinear periodic behavior)
• High-dimensional multiplicative test models (up to 300 variables)
• An engineering frame model with correlated inputs
• A structural reliability fatigue model (4R method)