Title :
A corrected version of KIC for robust model selection in small samples
Author :
Seghouane, Abd-Krim
Author_Institution :
Canberra Res. Lab., Nat. ICT Australia, Canberra, ACT
Abstract :
A small sample version of KIC for the selection of least absolute deviations regression models is proposed. In contrast to KIC, the proposed criterion named KICL1, where the notation L1 stands for absolute deviation, provides an exactly unbiased estimator for the expected Kullback symmetric divergence, assuming that the errors have a double exponential distribution and that the true model is correctly specified or overfitted. Simulation results showing that KICL1 performs slightly better than KIC are presented.
Keywords :
estimation theory; exponential distribution; regression analysis; signal sampling; Kullback information criterion; Kullback symmetric divergence; double exponential distribution; exactly unbiased estimator; least absolute deviation regression model; robust model selection; small KIC sample version; Australia Council; Computational modeling; Error correction; Exponential distribution; Gaussian distribution; Laboratories; Least squares methods; Parameter estimation; Parametric statistics; Robustness;
Conference_Titel :
Signal Processing and Its Applications, 2007. ISSPA 2007. 9th International Symposium on
Conference_Location :
Sharjah
Print_ISBN :
978-1-4244-0778-1
Electronic_ISBN :
978-1-4244-1779-8
DOI :
10.1109/ISSPA.2007.4555330
