CAREER: Methods for Data-Driven Service Engineering

Purdue University, West Lafayette IN

Investigators

Abstract

This Faculty Early Career Development Program (CAREER) project contributes to the advancement of the Nation's health, prosperity, and welfare by studying the fundamentals of data-driven engineering of service systems. Service industries such as healthcare, communications, transportation and logistics, comprised over 67 percent of US gross domestic product and employed more than 78 percent of the working population in 2020, according to the US Federal Reserve and the Bureau of Labor Statistics. Engineering service systems to increase operational efficiency and cost effectiveness is a crucial task with significant socio-economic implications for the US Designing and operating these systems has benefited from the use of sophisticated mathematical models of these systems. However, identifying a model appropriate to a system of interest is a remarkably hard problem. This award supports a fundamental understanding of how to utilize large operational data sets and emerging machine learning technologies, to facilitate such model identification. The research products of this award will enable engineers, policy makers and academicians to better improve operational and cost efficiencies, and as a consequence improve overall national welfare. The educational plan as part of this award aims to radically improve student advising and mentoring in large, public universities and colleges, by utilizing machine learning to predict student performance and enable student advisors and faculty mentors to implement early advising interventions. These interventions can substantially improve student retention and graduation rates, particularly among at-risk populations such as first-generation college students. Large service systems exhibit multi-scale spatiotemporal variations and they are naturally modeled using stochastic network models operating in random and nonstationary environments. Model identification and calibration are fundamental to data-driven engineering. This research project fills an important gap in the literature, where most of the focus is on stationary models. The research in this award proposes the development of novel semiparametric methods for model calibration in stochastic network models operating in random and nonstationary environments. The research leverages and extends state-of-art techniques from stochastic analysis, optimization over measure spaces and point process theory to develop these methods. In the case of model inference, the project investigates theory that builds on and significantly extends M-estimation, minimax and function approximation analyses to the semiparametric setting of service system models. The complex nature of stochastic network models creates a desideratum for an understanding of which classes of such models can be uniquely identified. The project addresses identifiability analyses and utilizes information geometry to understand the landscape or manifolds of the model space to address this question. Together, the theory and methods will lead to a comprehensive ‘graybox’ methodology for calibrating service system models. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.