Principle:Interpretml Interpret Morris Sensitivity Analysis

Metadata

Field	Value
Sources	Paper: Morris 1991 "Factorial sampling plans for preliminary computational experiments", Doc: SALib
Domains	Sensitivity_Analysis, Feature_Analysis
Updated	2026-02-07

Overview

A global sensitivity analysis method that screens input factors to identify which have important effects on model output using a one-at-a-time sampling design.

Description

Morris Sensitivity Analysis (Method of Morris) is a screening method that identifies which inputs are important, non-important, or involved in interactions, using a relatively small number of model evaluations. It computes two statistics per feature: μ* (mean of absolute elementary effects) measures overall importance, and σ (standard deviation of elementary effects) indicates the presence of interactions or non-linearities.

Usage

Use Morris sensitivity analysis when you have many features and want an efficient global screening of feature importance before deeper analysis. It is more computationally efficient than Sobol analysis.

Theoretical Basis

The elementary effect of the i-th factor:

$d_i(x)=\frac{f(x_1,...,x_i+\mathrm{\Delta },...,x_k)-f(x)}{\mathrm{\Delta }}$

${\mu }_i^{*}=\frac{1}{r}\sum _{j=1}^r|d_i^{(j)}|$

${\sigma }_i=\sqrt{\frac{1}{r}\sum _{j=1}^r(d_i^{(j)}-{\overline{d}}_i{)}^2}$

Related Pages

• Implementation:Interpretml_Interpret_MorrisSensitivity