An Approach to Variable Aggregation in Efficiency Analysis

• Author(s): Noncheva, Veska ; Mendes, Armando B. ; Silva, Emiliana
• Date: 2009
• Persistent ID: http://hdl.handle.net/10400.3/2144
• Origin: Repositório da Universidade dos Açores
• Subject(s): Canonical Correlation Analysis; Data Envelopment Analysis; Efficiency; Variable Aggregation; Multivariate Statistics; Data Mining
• Document Type: Article
• Language: English

Description

Conference: The paper is selected from International Conference "Classification, Forecasting, Data Mining" CFDM 2009, Varna, Bulgaria, June-July 2009. In the nonparametric framework of Data Envelopment Analysis the statistical properties of its estimators have been investigated and only asymptotic results are available. For DEA estimators results of practical use have been proved only for the case of one input and one output. However, in the real world problems the production process is usually well described by many variables. In this paper a machine learning approach to variable aggregation based on Canonical Correlation Analysis is presented. This approach is applied for efficiency estimation of all the farms in Terceira Island of the Azorean archipelago.