Hidden discriminative features extraction for supervised high-order time series modeling.

Area of Science:

• Data Science
• Machine Learning
• Signal Processing

Background:

• Time series data analysis often requires dimensionality reduction and feature extraction.
• Tensor-based methods offer advantages for multi-modal time series data.
• Existing methods may struggle with high-dimensional tensor data and small sample sizes.

Purpose of the Study:

• To introduce Tensor Discriminative Feature Extraction (TDFE), a novel orthogonal Tucker-decomposition-based method.
• To extract high-order discriminative subspaces from tensor-structured time series data.
• To enhance classification accuracy and data interpretability through supervised tensor modeling.

Main Methods:

• Employed orthogonal Tucker decomposition for feature extraction.
• Utilized category information to maximize between-class scatter and minimize within-class scatter.
• Applied the method to channel×frequency bin×time frame (EEG) and gene×sample×time (microarray) datasets.

Main Results:

• Achieved high classification accuracies: 98.26% for epilepsy EEG and 89.63% for microarray data.
• Demonstrated significant improvements over matrix-based and existing tensor-based discriminant decomposition approaches.
• Showcased reduced dimensionality, improved feature interpretability, and faster processing times.

Conclusions:

• TDFE effectively extracts discriminative features from tensor time series data.
• The method offers superior performance, especially with limited sample sizes.
• TDFE provides a robust and efficient approach for supervised tensor modeling and analysis.