1Computer Science and IT, School of Science, RMIT University, Melbourne, VIC, email@example.com
Context is the most influential signal in analysing human behaviours. Effective and efficient techniques for analyzing contexts inherent in the spatio-temporal sensor data from the urban environment are paramount, particularly in addressing these key growth areas in urbanization: human mobility, transportation, and energy consumption. It is important to observe and learn the context from which the data is generated in, particularly when dealing with heterogenous high-dimensional data from buildings, cities, and urban areas.
One main challenge in spatio-temporal analytics is to discover meaningful correlations among the numerous sensor channels and other types of data from multiple domains. Often big data is not the problem, but sparse data is. High quality annotations required are often not available. Another major issue is the dynamic changes in the real-world, requiring a model robust to the fast-changing urban environment. I will present our generic temporal segmentation techniques that we have used for multiple applications. I will then present the applicability of some of our ensemble methods for multivariate and multi-target prediction in real-world cases, such as parking violation monitoring, predicting daily trajectories, visitor behaviour analysis, transport mode and activity recognition, and crime prediction. A new concept of cyber, physical, social contexts will be introduced, and how they translate in various domain applications of our research for smarter cities and smarter buildings, and intelligent assistants.