Regularized Information Loss for Improved Model Selection
Date
2023
Journal Title
Journal ISSN
Volume Title
Publisher
Springer Science and Business Media Deutschland GmbH
Abstract
Information criteria are used in many applications including statistical model selection and intelligent systems. The traditional information criteria such as the Akaike information criterion (AIC) do not always provide an adequate penalty on the number of model covariates. To address this issue, we propose a novel method for evaluating statistical models based on information criterion. The proposed method, called regularized information criterion (RIL), modifies the penalty term in AIC to reduce model overfitting. The results of numerical experiments show that RIL provides a better reflection of model predictive error than AIC. Thus, RIL can be a useful tool in model selection. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
Description
item.page.type
Book chapter
item.page.format
Keywords
Akaike information criterion, Computational statistics, Information criteria, Intelligent systems, Model selection
Citation
Kamalov, F., Moussa, S. & Reyes, J.A. (2023). Regularized Information Loss for Improved Model Selection. In G. Rajakumar, KL Du, & A. Rocha (Eds.) Intelligent Communication Technologies and Virtual Mobile Networks. ICICV 2023. Lecture Notes on Data Engineering and Communications Technologies, 171 (pp. 801 – 811). Springer, Singapore. https://doi.org/10.1007/978-981-99-1767-9_58