Enhancing data envelopment analysis: a comprehensive calibration with random forests

Table 1 Monte Carlo results: MSE of scenarios with \(10\%\) outliers

n	P	RF-DEA	DEA	Imp\(\%\)	RF-DEA	DEA	Imp\(\%\)
n	P	Q = 1			Q = 2
n=50	p=2	0.094 (0.012)	0.113 (0.019)	16.88%	0.103 (0.010)	0.108 (0.014)	4.58%
n=50	p=3	0.106 (0.010)	0.116 (0.014)	8.45%	0.114 (0.009)	0.117 (0.011)	3.11%
n=100	p=2	0.083 (0.009)	0.127 (0.020)	34.59%	0.096 (0.008)	0.120 (0.016)	20.15%
n=100	p=3	0.096 (0.007)	0.116 (0.014)	17.20%	0.105 (0.008)	0.115 (0.012)	9.13%
n=200	p=2	0.079 (0.006)	0.146 (0.016)	45.75%	0.086 (0.006)	0.132 (0.014)	34.66%
n=200	p=3	0.088 (0.006)	0.130 (0.013)	32.40%	0.095 (0.005)	0.122 (0.010)	21.74%
n=400	p=2	0.076 (0.005)	0.163 (0.010)	53.47%	0.080 (0.005)	0.145 (0.011)	44.48%
n=400	p=3	0.082 (0.005)	0.145 (0.010)	43.49%	0.088 (0.005)	0.133 (0.010)	33.85%

Notes: Imp%: Improvement in performance (e.g., \(1-\text {MSE}_{\text {RF-DEA}}/\text {MSE}_{\text {DEA}}\))
Standard deviation in parentheses
Number of replications = 100, VRS assumption