Joint modeling of multiple outcomes over space and time (JMMOST): A Bayesian approach

点击翻译按钮获取中文摘要

The Big Data market is currently valued at 140 billion USD and, by 2025, is expected to hold about 175 Zettabytes (1 ZB = 1e12 GB) of data that will require regular processing and analysis for information extraction. The proposed research addresses this emerging challenge of analyzing big data, more specifically rich data, which are cleaned, processed, and refined forms of raw-big data. It focuses on developing novel multivariate analytical methods that can analyze the precise location and time information in rich data, providing knowledge about events that are or will be happening around us at any point in time. For example, analyzing rich crime data at space-time (ST) dimensions can help detect the hourly, daily, or weekly progression of different crime types in neighborhoods. Thus, allowing targeted crime monitoring and the best use of our finite resources. The project will innovate the joint modeling of multiple outcomes over space and time (JMMOST) to analyze multiple complex outcomes (or events) from rich ST data using a single model for generating new information hidden in rich data. However, integrating space and time dimensions of data in a single model can be highly challenging due to their contrasting nature, which is further complicated by the large volume of rich ST data. These analytical constraints will be addressed through the application of Bayesian spatiotemporal modeling at a small-area level (e.g., neighborhoods). The selection of the Bayesian framework is based on past research evidence that Bayesian techniques supersede conventional approaches in analyzing space and time data. The project has short-term goals that will help us achieve our long-term goals in perfecting JMMOST for rich data analysis and establishing a Rich Data Spatial Analysis Research Centre in Canada that supports Bayesian spatial and ST analysis. The short-term goals aim to develop novel JMMOST methods for analyzing rich data at a fine--scale of space (e.g., customized grids) and time (e.g., hours), which can better capture the ST variabilities in rich data. The long-term goals aim to perfect the novel methods in terms of robustness and flexibility and thus, allow real-time analysis of rich data with spatial and temporal components for surveillance systems. This is important because, without the availability of reliable methodologies for analyzing rich data, the progress in natural science and engineering (NSE) research in Canada could stall due to the failure to exploit the ever-growing rich ST data sources. The project will globally benefit NSE fields like geographic information science and other fields like criminology and economics, enabling them to analyze their rich data to obtain new information. Through my current NSERC grant, I have established a research team to complete the proposed research. Upon renewal of the grant, more HQPs will be trained to develop and apply JMMOST methods in different fields using Canadian rich data.