Evaluation Criterion of Linear Model Order Selection Approaches Based Average Kullback-Leibler Divergence
Yuming Du · 2009
Average Kullback-Leibler divergence (AKD) between the selected model and the true model is proposed as an available measurement for evaluating different model order selection approaches in simulations. Kullback-Leibler divergence of linear model order is reduced to simple forms, so AKD of linear model can be easily computed. In terms of parameter estimation of linear model, simulation results show that the AKD is a more reasonable measurement than naive methods.