ConceptExplorer: Visual Analysis of Concept Drifts in Multi-source Time-series Data

Xumeng Wang, Wei Chen, Jiazhi Xia, Zexian Chen, Dongshi Xu, Xiangyang Wu, Mingliang Xu, Tobias Schreck

External link (DOI)
View presentation:2020-10-27T19:00:00ZGMT-0600Change your timezone on the schedule page
2020-10-27T19:00:00Z
Exemplar figure
It is challenging to follow the evolution of complex correlations in time-series data. To solve this problem, we first detect the accuracy drops of the prediction model as significant changes. The patterns in the time segments between two changes can be considered consistent. Related data records can, therefore, be integrated for analysis.
Fast forward

Direct link to video on YouTube: https://youtu.be/KqB3Gy1eHvQ

Keywords

Temporal data, data analysis, reasoning, problem solving, and decision making, machine learning techniques.

Abstract

Time-series data is widely studied in various scenarios, like weather forecast, stock market, customer behavior analysis. To comprehensively learn about the dynamic environments, it is necessary to comprehend features from multiple data sources. This paper proposes a novel visual analysis approach for detecting and analyzing concept drifts from multi-sourced time-series. We propose a visual detection scheme for discovering concept drifts from multiple sourced time-series based on prediction models. We design a drift level index to depict the dynamics, and a consistency judgment model to justify whether the concept drifts from various sources are consistent. Our integrated visual interface, ConceptExplorer, facilitates visual exploration, extraction, understanding, and comparison of concepts and concept drifts from multi-source time-series data. We conduct three case studies and expert interviews to verify the effectiveness of our approach.