Human-in-the-Loop: Visual Analytics for Building Models Recognizing Behavioral Patterns in Time Series

Area of Science:

• Data Science
• Computer Science
• Human-Computer Interaction

Background:

• Detecting complex behavioral patterns in temporal data is challenging due to imprecise specifications and noise sensitivity.
• Traditional methods struggle with noisy data, while machine learning requires extensive labeled datasets.
• Existing approaches often fail to capture subtle patterns or discover unexpected behaviors effectively.

Purpose of the Study:

• To develop a visual analytics framework for deriving, testing, and combining interval-based features for pattern discrimination.
• To enable domain experts to generate training data for machine learning algorithms.
• To enhance the recognition and characterization of both expected and unexpected patterns in temporal data.

Main Methods:

• A visual analytics approach empowering domain experts to define and refine pattern features.
• Integration of user-driven feature engineering with machine learning model training.
• Utilizing visual aids for pattern recognition, characterization, and discovery.

Main Results:

• Demonstrated feasibility and effectiveness through case studies.
• Improved accuracy in detecting complex behavioral patterns in temporal data.
• Successful generation of training data for machine learning algorithms by domain experts.

Conclusions:

• The visual analytics approach offers a novel framework for integrating human expertise with machine learning.
• This method advances data analytics by improving the detection of behavioral patterns in temporal data.
• It provides a practical solution for challenges in pattern recognition and machine learning data preparation.