The clustering of unbounded data-streams is a difficult problem since the observed instances cannot be stored for future clustering decisions. Moreover, the probability distribution of streams tends to change over time, making it challenging to differentiate between a concept-drift and an anomaly. Although many excellent data-stream clustering algorithms have been proposed in the past, they are not suitable for capturing the temporal contexts of an entity.

In this paper, we propose pcStream; a novel data-stream clustering algorithm for dynamically detecting and managing sequential temporal contexts. pcStream takes into account the properties of sensor-fused data-streams in order to accurately infer the present concept, and dynamically detect new contexts as they occur. Moreover, the algorithm is capable of detecting point anomalies and can operate with high velocity data-streams. Lastly, we show in our evaluation that pcStream outperforms state-of-the-art stream clustering algorithms in detecting real world contexts from sensor-fused datasets. We also show how pcStream can be used as an analysis tool for contextual sensor streams.