Skip to main content

Table 1 Summary of some characteristics of VIMs

From: Intervention in prediction measure: a new approach to assessing variable importance for random forests

Methods

GVIM

PVIM (CART-RF)

PVIM (CIT-RF)

CPVIM

varSelRF

varSelMD

IPM

Main references

[43]

[2]

[13, 15]

[25]

[26]

[10, 33]

[11] and this manuscript

Key characteristic

Node impurity

Accuracy after variable permutation

Accuracy after variable permutation

Alternative of PVIM; Conditional permutation

Backward elimination

Variable selection based on MD

Variables intervening in prediction

RF-based

CART-RF

CART-RF

CIT-RF

CIT-RF

CART-RF

CART-RF

CART-RF or CIT-RF

Handling of response

Univariate

Univariate

Univariate

Univariate

Categorical

All (multivariate included)

All (multivariate included)

Main R implementation

randomForest [18, 19]

randomForest [18, 19] randomForestSRC [20–22]

party [23–25]

party [23–25]

varSelRF [27, 28]

randomForestSRC [20–22]

Additional file 2

Casewise importance

No

Yes

Not defined

Not defined

No

No

Yes