A Divergence-Based Method for Weighting and Averaging Model Predictions
Provides a general, theoretically motivated model averaging method that improves prediction accuracy for practitioners using ensemble methods, particularly in small-sample settings.
This paper introduces a minimum divergence framework for weighting and averaging probabilistic predictions from multiple models, showing empirically that it outperforms or matches standard methods like stacking and AIC-based weighting, especially with small sample sizes.
This paper uses a minimum divergence framework to introduce a new way of calculating model weights that can be used to average probabilistic predictions from statistical and machine learning models. The method is general and can be applied regardless of whether the models under consideration are fit to data using frequentist, Bayesian, or some other fitting method. The proposed method is motivated in two different ways and is shown empirically to perform better than or on a par with standard model averaging methods, including model stacking and model averaging that relies on Akaike-style negative exponentiated model weighting, especially when the sample size is small. Our theoretical analysis explains why the method has a small-sample advantage.