| The widespread adoption of the internet of things is producing a huge amount of spatiotemporal data. Global mobile data traffic is growing an average of 60 percent and has reached 7.2 exabytes per month in 2016. Realizing values from these data promises to improve understanding of human activities that will up opportunities to innovative applications and services. To this end, my dissertation focuses on analyzing massive spatiotemporal data to gather actionable intelligence using machine learning techniques. I strive to address the challenges in processing, representing and mining spatiotemporal data on a large-scale by embracing big data analytics. In particular, this dissertation consists of three essays of predictive modeling. The first essay proposes a novel model that captures spatial dependencies, temporal preferences, and social influence to predict when and where people will use certain kinds of services. In the second essay, I demonstrate the ability of network analysis and sequence learning to glean previously unavailable insights for student retention prediction, which fill the gap between student attrition theories and existing quantitative approaches. Using public transportation as a proxy, the third essay presents a deep-learning-based urban mobility model that collectively forecast short-term ridership between each pair of urban regions. Throughout this thesis, I show how big data infrastructure such as Hadoop MapReduce can amplify the capabilities to analyze spatiotemporal data in an efficient and economical way. All the proposed models are developed and evaluated on real-world datasets. In conjunction with prior knowledge in the application domains, the results can be converted into actionable intelligence that supports decision making in operation management. |
