Experimental Analysis of Moments of Predictive Deviations as Ensemble Diversity Measures for Model Selection in Time Series Prediction

This paper presents an experimental analysis of moments of predictive deviations as measures of ensemble diversity to estimate the performance of time series prediction for model selection. As an extension of the ambiguity decomposition of bagging ensemble, we decompose the fourth power of ensemble prediction error and examine the effect of the moments of predictive deviations of ensemble members to the ensemble prediction error. By means of numerical experiments, we analyze the results to show the properties and the effectiveness of the moments.