Cloud Model-Based Adaptive Time-Series Information Granulation Algorithm and Its Similarity Measurement.

Area of Science:

• Data Mining
• Time Series Analysis
• Artificial Intelligence

Background:

• Dimensionality reduction is crucial for efficient time series data mining.
• Existing methods may require pre-specification of parameters, limiting adaptability.
• Cloud model theory offers a novel approach for representing uncertainty and vagueness.

Purpose of the Study:

• To propose a novel information granulation method for time series dimensionality reduction using cloud model theory.
• To develop a corresponding similarity measurement for granular time series.
• To evaluate the effectiveness of the proposed methods in time series clustering.

Main Methods:

• An information granulation validity index (IGV) based on cloud model entropy and expectation was developed.
• An adaptive information granulation algorithm for time series (CMAIG) was proposed, transforming time series into granular representations without pre-specifying granule number.
• A novel similarity measurement method (CMAIG_ECM) was designed for granular time series, integrated into a hierarchical clustering algorithm (CMAIG_ECM_HC).

Main Results:

• The CMAIG algorithm achieved efficient dimensionality reduction by transforming time series into granular representations (normal clouds).
• The CMAIG_ECM similarity measurement effectively captured relationships between granular time series.
• Experiments on UCR and stock datasets showed superior clustering performance of CMAIG_ECM_HC across various time series shapes and trends.

Conclusions:

• The proposed cloud model-based information granulation and similarity measurement offer an effective approach for time series dimensionality reduction and clustering.
• CMAIG_ECM_HC demonstrates robust performance, outperforming existing methods on diverse time series datasets.
• This work advances time series data mining by providing an adaptive and efficient granulation technique.