Visual discovery and model-driven explanation of time series patterns

摘要

Gatherminer is an interactive visual tool for analysing time series data with two key strengths. First, it facilitates bottom-up analysis, i.e., the detection of trends and patterns whose shapes are not known beforehand. Second, it integrates data mining algorithms to explain such patterns in terms of the time series' metadata attributes - an extremely difficult task if the space of attribute-value combinations is large. To accomplish these aims, Gatherminer automatically rearranges the data to visually expose patterns and clusters, whereupon users can select those groups they deem `interesting.' To explain the selected patterns, the visualisation is tightly coupled with automated classification techniques, such as decision tree learning. We present a brief evaluation with telecommunications experts comparing our tool against their current commercial solution, and conclude that Gatherminer significantly improves both the completeness of analyses as well as analysts' confidence therein.