TrSAX-An improved time series symbolic representation for classification.

Area of Science:

• Data Mining
• Machine Learning
• Time Series Analysis

Background:

• Symbolic Aggregate approXimation (SAX) is a common method for time series data mining.
• SAX represents time series segments using mean values, but overlooks trend information.
• This limitation can lead to misclassification when time series have similar means but different trends.

Purpose of the Study:

• To propose an improved symbolic representation for time series data.
• To enhance the accuracy of time series classification by capturing both mean and trend information.
• To address the limitations of the standard SAX method in distinguishing time series with varying trends.

Main Methods:

• Developed an enhanced symbolic representation by integrating the least squares method with SAX.
• The new representation captures both the mean value and the trend of time series segments.
• Evaluated the proposed method by comparing its classification performance against original SAX and other competitive classifiers for short time series.

Main Results:

• The proposed integrated SAX representation demonstrated a lower error rate compared to the original SAX.
• Classifiers utilizing the enhanced representation outperformed five other representative and competitive classifiers on most datasets.
• The integration effectively captures crucial trend information previously ignored by SAX.

Conclusions:

• The proposed method offers a more robust symbolic representation for time series data.
• Incorporating trend information alongside mean values significantly improves classification accuracy.
• This enhanced SAX approach is a valuable advancement for time series data mining and analysis.