SIMIT: Subjectively Interesting Motifs in Time Series

Area of Science:

• Data Science
• Machine Learning
• Time Series Analysis

Background:

• Numerical time series data are abundant across various fields, including healthcare and industry.
• Identifying recurring approximate subsequences (motifs) within time series is crucial for extracting meaningful information.
• Existing methods face challenges in defining and efficiently discovering subjectively interesting motifs.

Purpose of the Study:

• To formalize the notion of motif interestingness in time series data.
• To develop a subjective, information-theoretic measure for quantifying motif interestingness.
• To devise an efficient computational approach for mining interesting motifs.

Main Methods:

• Formalized motif definition and an information-theoretic framework for subjective interestingness.
• Developed relaxations and a branch-and-bound algorithm.
• Implemented the approach using a constraint programming solver.

Main Results:

• Successfully quantified motif interestingness based on user-specific prior expectations.
• Demonstrated the ability to mine subjectively interesting motifs in time series.
• Experimental validation on synthetic and real-world datasets confirmed the approach's efficacy.

Conclusions:

• The proposed method provides a novel way to define and discover user-relevant motifs in numerical time series.
• The approach addresses key challenges in motif interestingness formalization and computational tractability.
• Enables the identification of meaningful patterns in small to mid-sized time series datasets.