Kolloquium Geoinformatik Predictive Analytics and Knowledge Reasoning for Spatial Data: A Case Study of China Smog

Predictive analytics like time-series forecasting plays an important role in big spatial data analysis. This report first presents a case study of big spatial data analysis about China smog (big air pollution event), which adopts multiple physical sensor data streams and social media (e.g., Weibo) data streams. Different data mining techniques (e.g., feature extraction, ensemble learning, artificial neural network, text analysis) are adopted and evaluated for the prediction problems in smog and smog-related health analysis. Good results can be achieved with a suitable combination of these algorithms. However, such traditional data mining still has big improvement space for spatio-temporal prediction with big data. Machine learning models may encounter the problems in solving concept shift, providing explanations, fusing big data, etc. This report also presents my study of consistent prediction method, which incorporates knowledge reasoning for the concept shift problem in supervised stream learning with the case study of air quality forecasting. In this part, my personal ideas of why we need knowledge graph (i.e., linked data) besides raw data records and how integrating learning and reasoning benefits predictive analytics of spatial data will be briefly presented.