Mining candidate causal relationships in movement patterns

Abstract

In many applications, the environmental context for and drivers of movement patterns are just as important as the patterns themselves. This article adapts standard data mining techniques, combined with a foundational ontology of causation, with the objective of helping domain experts identify candidate causal relationships between movement patterns and their environmental context. In addition to data about movement and its dynamic environmental context, our approach requires as input definitions of the states and events of interest. The technique outputs causal and causal-like relationships of potential interest, along with associated measures of support and confidence. As a validation of our approach, the analysis is applied to real data about fish movement in the Murray River in Australia. The results demonstrate that the technique is capable of identifying statistically significant patterns of movement indicative of causal and causal-like relationships.

Keywords:

• movement patterns
• context-aware movement analysis
• sequence mining
• causation
• geosensor networks
• environmental monitoring

Acknowledgements

This research was partly funded by the Australian Research Council Discovery Project DP120100072 ‘From environmental monitoring to management: Extracting knowledge about environmental events from sensor data’.

Notes

1. 1. Note that the usual range of options exist for thresholding continuous data into qualitative categories, including equal interval, quantiles, k-means, and so forth.

2. 2. In fact, a surprisingly small number of fish in the study ever move far. Of the 1050 tagged fish, only around 260 are ever recorded moving between river zones. Consequently, the support and confidence reported arguably underestimates the strength of causal relationships; although only 4% of fish engage in rapid upstream movement (E₄/E₆), this translates to approximately 17% of the fish that ever move.