Enhancing Pattern Detection in Movement Data: An Interactive Approach with Domain Expertise Integration

Using interactive visualisation, we explored the performance of a knowledge-based system designed to detect complex activity patterns in vessel movement using formal specifications obtained from domain experts. We found that the system does not perform well enough, and we identify the primary cause as the treatment of human-given definitions as precise and strict constraints, whereas human reasoning is flexible and mostly approximate. To address this issue, we propose a novel visual analytics approach that leverages expert knowledge and formal logical inference results for constructing features that effectively differentiate patterns of interest from other types of movement. These features are then employed for interactive classification of a subset of movement episodes, which creates labelled examples of pattern types to be utilised to build a feature-based pattern classifier. To evaluate our approach, we conducted a case study focusing on the detection of trawling activities in trajectories of fishing vessels. The results of our study demonstrate a significant improvement in pattern recognition, highlighting the enhanced flexibility in applying domain knowledge. Our contribution to the field of movement analytics is a novel framework that integrates human expertise and analytical reasoning with ML or AI techniques. By bridging the gap between human reasoning and formal inference, our framework improves the detection of complex behavioural patterns in movement data.