A practical utility-based but objective approach to model selection for regression in scientific applications

• DOI: 10.1007/s10462-023-10591-4
• Publication Date: 2023-10-05T06:01:36Z
• Authors: Andrea Murari; Riccardo Rossi; Luca Spolladore; Michele Lungaroni; Pasquale Gaudio; Michela Gelfusa
• Keywords: Generality; Akaike information criterion; Bayesian information criterion

Abstract In many fields of science, various types models are available to describe phenomena, observations and the results experiments. last decades, given enormous advances information gathering technologies, also machine learning techniques have been systematically deployed extract from large databases. However, regardless their origins, no universal criterion has found so far select most appropriate model data. A unique solution is probably a chimera, particularly in applications involving complex systems. Consequently, this work utility-based approach advocated. solutions proposed not purely subjective but all based on “objective” criteria, rooted properties data, preserve generality allow comparative assessments results. Several methods developed tested, improve discrimination capability basic Bayesian theoretic with particular attention BIC (Bayesian Information Criterion) AIC (Akaike indicators. Both quality fits evaluation complexity aspects addressed by proposed. The competitive advantages individual alternatives, for both cross sectional data time series, clearly identified, together application. improvements criteria selecting right more reliably, efficiently terms requirements can be adjusted very different circumstances applications. Particular paid ensure that versions indicators easy implement practice, confirmatory exploratory settings. Extensive numerical tests performed support conceptual theoretical considerations.