. Author manuscript; available in PMC: 2023 Jan 1.

Published in final edited form as: J Comput Graph Stat. 2021 Nov 17;31(2):390–402. doi: 10.1080/10618600.2021.1987253

Table 3:

Average prediction error of the avalanche survival models for each method (LEFT) and variable importance of the ICRF model fitted on the first training set of the avalanche data (RIGHT). ICRF (Q), quasi-honest ICRF; ICRF (E), exploitative ICRF; The importance values are rescaled so that maximum values for each measure becomes 1. The multiplier is the original importance scale.

Prediction error			Variable importance
method	IMSE₁ (sd)	IMSE₂ (sd)	variable	quasi-honest	exploitative
ICRF (Q)	0.026 (0.0032)	0.026 (0.0038)	by IMSE₁ (multiplier = 0.0073)
ICRF (E)	0.022 (0.0021)	0.019 (0.0013)	Burial depth	1.00	0.47
Fu	0.024 (0.0030)	0.027 (0.0042)	Group activity	0.17	0.41
Fu (*)	0.023 (0.0028)	0.020 (0.0032)	Location	0.16	0.24
Yao	0.025 (0.0031)	0.026 (0.0031)	by IMSE₂ (multiplier = 0.0041)
Yao (*)	0.026 (0.0030)	0.026 (0.0030)	Burial depth	1.00	0.67
Cox	0.021 (0.0025)	0.019 (0.0021)	Group activity	0.27	0.46
Cox (*)	0.021 (0.0026)	0.019 (0.0022)	Location	0.55	0.27