Time-Sensitive Feature Mining for Temporal Sequence Classification

Abstract

Behavior analysis received much attention in recent year, such as customer-relationship management, social security surveillance and e-business. Discovering high impact-driven behavior patterns is important for detecting and preventing their occurrences and reducing resulting risks and losses to our society. In data mining community, researchers pay little attention to time-stamps in temporal behavior sequences (without explicitly considering inherent temporal information) during classification. In this paper, we propose a novel Temporal Feature Extraction Method - TFEM. It extracts sequential pattern features where each transition is annotated with a typical transition time (its duration or interval). Therefore it substantially enriches temporal characteristics derived from temporal sequences, yielding improvements in performances, as demonstrated by a set of experiments performed on synthetic and real-world datasets. In addition, TFEM has the merit of simplicity in implementation and its pattern-based architecture can generate human-readable results and supply clear interpretability to users. Meanwhile, it is adjustable and adaptive to user’s different configurations, allowing a tradeoff between classification accuracy and time cost.