Publication Date
2010-03-01
Availability
Open access
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PHD)
Department
Industrial Engineering (Engineering)
Date of Defense
May 2008
First Committee Member
Dr. Sohyung Cho - Committee Chair
Second Committee Member
Dr. Murat Erkoc - Committee Member
Third Committee Member
Dr. Shihab Asfour - Committee Member
Fourth Committee Member
Dr. Moiez Tapia - Outside Committee Member
Abstract
Data envelopment analysis (DEA) method which is based on a mathematical programming approach, and stochastic frontier functions (SFF) which is based on the econometric regression approach are two well-known tools for performance and efficiency analysis for profit and non-profit organizations, called decision making units (DMUs). While SFF accounts for both managerial and observational errors, DEA assumes that all of the errors are due to only managerial errors, which can be misleading to decision-makers and managers, if the data utilized is contaminated with statistical noise. The challenge therefore facing empirical or traditional DEA?s methodology, is how to account for both managerial and observational errors if present in the analysis, so as to determine DEA?s ?true? or optimal frontiers. The main objective of this dissertation is to determine DEA?s "true" frontier in a totally nonparametric environment, by utilizing traditional DEA efficient frontiers, along with DEA inefficient frontiers. DEA is integrated with SFF, thus enabling the identification of efficient frontiers, and specifically, a machine learning technique called support vector machine (SVM) is employed to provide an adaptive way to estimate ?true? frontiers for a set of input-output data, considering both managerial and observational errors/deviations. A ratio based on statistics for managerial and observational errors is utilized to find the ?true? frontiers that perform in between two extremes, and the methodology developed is applied to a real data set where frontiers generated by SVM are compared to ones obtained by the neural network (NN), and ordinary least squares (OLS) regression approaches. The results showed that SVM outperformed NN and OLS regression by about 2-to-1 in estimating nonlinear functions for efficient and inefficient frontiers. Also, utlizing a ratio based on statistics for managerial and observational errors, SVM gave a better estimation of the ?true? frontier for DEA than both NN and OLS. The work in this research can prevent managers and decision-makers from committing grievous errors relative to the allocation and distribution of the funds and resources of their organizations, as well as, help organizations to plan a more realistic investment by providing reasonable sense of benchmarking to their peers (DMUs).
Keywords
True Frontiers; Managerial Errors; Efficient Frontiers; Inefficient Frontiers; DEA Errors; Observational Errors
Recommended Citation
Poitier, Kerry R., "True Efficient Frontiers in Production Performance Assessment Using Machine Learning and Data Envelopment Analysis" (2010). Open Access Dissertations. 927.
https://scholarlyrepository.miami.edu/oa_dissertations/927