Classification of Event Sequences Based on Temporal Relation Features

Temporal data include not only time-stamped raw data but also time intervals for events with a non-zero duration. Classification of interval-based event sequences has been an active research topic in the data science community. One major research issue in sequence classification is to extract discriminative features that properly capture the underlying sequences for high accuracy classification. Previous research on sequence classification mainly focused on time series using time-point based features. In this paper, we propose to define features based on Allen’s temporal relations between time intervals. Based on our earlier work on temporal data modeling, we develop a novel scheme for sequence representation of event-intervals. We describe the detailed algorithms and report the experimental results.