DENSITY BASED FRAMEWORKS FOR SPATIO-TEMPORAL CHANGE ANALYSIS

Analyzing change in spatial data is important for many different domains such as biology, ecology, meteorology, medicine, transportation, and forestry. On the other hand, density functions have served as a valuable tool in data mining. However, the development of spatiotemporal data mining techniques based on density estimation functions, particularly for change analysis, is still in its infancy. To address the need for more robust and versatile data analysis tools, density-based frameworks for spatio-temporal change analysis are designed and implemented in this dissertation. As a part of this research, we introduce density-based spatio temporal data analysis, change analysis, and storytelling frameworks for emotion mapping and emotion change analysis. Our proposed approach first segments first the input tweet dataset into batches based on a fixed-size time window. Next, by generalizing existing kernel density estimation techniques, each batch is transformed into a weighted continuous function that takes positive and negative values. After that, we employ a contouring algorithm to obtain spatial clusters. Finally, an emotion change graph whose nodes are the spatial clusters and whose edges capture the temporal relationships of different types between spatial clusters is constructed and mined based on story types. The frameworks were successfully applied to tweets collected in the state of New York in June 2014. The experimental results show that the framework can effectively discover interesting spatiotemporal patterns in tweets and analyzes how emotions change over time. A second theme of this research is the development of density-based collocation mining frameworks. As density functions have been demonstrated to be an efficient tool to model the distance-decay effect, we developed new density-based collocation measures estimating the strength of a given collocation pattern in a particular location. This approach is generalized to 3D spatio-temporal space to mine spatio-temporal collocation patterns using the 3D space-time density functions and a batch-based approach. Our experimental evaluation using the NYC real-world crime data set has demonstrated that our approach provides more insights compared to the traditional collocation methods. Finally, as polygons play an important role in spatio-temporal data analysis, we extend our first approach to support anomaly detection in polygon data sets. A novel framework to analyze anomalous drought changes is introduced based on the Subdue system. The experiments tested the use of the proposed framework on a time-series of U.S drought polygons between 2015 and 2021.