A Novel Variable Selection Approach Based on Multi-criteria Decision Analysis

• Published in: Information Processing and Management of Uncertainty in Knowledge-Based Systems Vol. 1602; pp. 115 - 127
• Main Authors: Xie, Shengkun, Zhang, Jin
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2022; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Complex systems; Machine learning; TOPSIS; Variable importance measure; Variable selection
• ISBN: 9783031089732; 3031089731
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-031-08974-9_9

Real-world complex systems, such as transportation and insurance systems, have constantly produced massive data, and the variables used to capture their variability may be overwhelming. Therefore, it is important to balance the model’s interpretability and prediction accuracy when building a predictive model. Keeping a balance on these two aspects may significantly improve prediction reliability and maintain key knowledge or information from complex systems so that the overall control and management of the complex system are statistically optimal. The paper proposes a novel approach for variable selection based on the importance measures from different data sources or different types of measures of importance obtained from machine learning models. The method formulates the variable selection problem in terms of multi-criteria decision analysis. It aims to bring a systematic way for decision-making in terms of variable selection to build more interpretable predictive models.