RII Track-4: Harnessing Big Event Data with Heterogeneous Feature: Intelligent Food-Borne Outbreak Investigations and Beyond

University Of Arkansas, Fayetteville AR

Investigators

Abstract

This award will advance the Nation's food safety, cyber security and economic welfare by innovating new statistical learning methodologies that enhance the critical capabilities of harnessing big event data with heterogeneous feature information. The penetration of Big Data technologies into interdisciplinary domains has led to an explosive growth of large-scale recurrent event data. Examples include, but not limited to, recurrent outbreaks of food-borne diseases from urban blocks in major cities, repeated cyber-attacks against vital infrastructures, recurring disasters or extreme weather events at critical geo-locations, and failures experienced by repairable engineering systems under dynamic operating-environmental conditions. This award will investigate the integration of modern additive-tree-based statistical learning approaches and classical point processes models for the modeling, prediction and optimization of large-scale recurrent event processes. Through the collaboration with the Industrial and Applied Genomics team at IBM Almaden Research Center (San Jose, CA), this project will test and validate the capabilities of the proposed methodologies in accelerating food-borne outbreak investigations using real data. The project also includes activities to benefit the PI's home institution, including nation-wide competitive intern programs that are currently rare for students at the PI's jurisdiction, especially for underrepresented communities. Integrating the research outcomes into the Data Analytics Minor program at the PI's home institution will nurture a pool of next-generation data scientists and engineers for the northwest Arkansas. This project will investigate a set of new additive-tree-based statistical learning methods to enable effective modeling, prediction and optimization of large-scale event processes with heterogeneous feature information. Based on the actual use cases provided by IBM research teams, this project will extensively investigate and demonstrate the advantages of a promising idea that integrates modern additive-tree-based methods and classical statistical models for stochastic point processes. This project consists of three Research Tasks (RT) during the PI's visit to IBM Almaden Research Center. RT1 and RT2 will propose two algorithms, RF-E and Boost-E, for modeling large-scale event data with both static and dynamic features. The two algorithms are deeply rooted in the framework of Random Forests and Gradient Boosting, respectively. RT3 will perform comprehensive model testing and validation on intelligent food-borne outbreaks investigation, with the critical support from IBM Research. Due to the interdisciplinary nature of the proposed research, the developed methodologies will lead to innovative solutions for a spectrum of event analytics applications arising from cross-disciplinary domains, including food safety, cyber security, reliability, online retail, transportation safety, disaster and extreme weather events. 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.