Skip to main content

Table 2 Comparison of accuracy and bias metrics for predictive models used

From: Comparing regression modeling strategies for predicting hometime

Model Root Mean Square Error Mean Absolute Error Bias Minimum Predicted Value Maximum Predicted Value Calibration Slope
Statistical Methods
 Linear Regression 28.82 24.13 -0.26 -53.74 103.37 1.00
 Ordinal Logistic Regression 28.64 23.96 -0.38 0.23 84.03 1.04
 Poisson Regression 29.02 24.50 -0.25 2.90 144.98 0.95
 Negative Binomial Regression 30.15 25.15 0.75 2.47 189.83 0.77
 Zero Inflated Poisson Regression 28.47 23.68 -0.31 0.17 95.59 1.04
 Zero Inflated Negative Binomial Regression 28.53 23.74 -0.31 0.18 97.46 1.03
 Cox Proportional Hazards Model 29.29 25.62 -1.64 0.00 77.00 1.33
 Hurdle Regression
(negative binomial zero distribution, Poisson distribution)
28.47 23.65 -0.25 0.50 95.99 1.02
Machine Learning Methods
 Random Forests Regression 28.32 23.08 -0.40 0.04 85.83 0.98
 Bagged Regression Trees 29.48 24.98 -0.25 18.20 73.29 1.06
 Support Vector Regression 29.18 21.55 2.08 -17.91 91.99 0.74
 Generalized Boosting Machine (Gaussian Distribution, Interaction Depth = 2) 28.39 23.89 -0.30 3.23 78.72 1.11
 Generalized Boosting Machine (Poisson Distribution, Interaction Depth = 15) 27.89 22.81 -0.35 3.49 83.39 1.01
 Lasso Regression 28.82 24.14 -0.26 -53.45 103.21 1.00
 Ridge Regression 28.83 24.25 -0.27 -50.06 101.93 1.03
  1. *a plausible minimum predicted value is ≥ 0, a plausible maximum predicted value is ≤ 90