A Modified Information Criterion for Model Selection

Abstract

Information criterion is an essential measure in data analysis. Primarily, information criterion is used to choose the statistical models. Because of that role, the development of the criteria becomes very crucial issue. In this study, a modified version of Fisher information criterion (FIC) is proposed to improve the classical FIC. Shrinkage estimation is adopted within FIC and also an additional penalty term is added multiplicatively. Suggested criterion is experienced on Lasso regression. The performance of the modified FIC is illustrated on simulated and real data sets. Empirical evidences demonstrate the success of the modified version of FIC for model selection when comparing with traditional criteria. © 2019 Taylor & Francis Group, LLC.