Random projection ensemble classification with high-dimensional time series.

Area of Science:

• Machine Learning
• Data Science
• Statistics

Background:

• Multivariate time-series (MTS) data are common across many fields.
• High-dimensional MTS data present significant analytical challenges.
• Existing classification methods struggle with the complexity of high-dimensional MTS.

Purpose of the Study:

• To develop novel random projection ensemble classifiers for high-dimensional MTS data.
• To address the challenges of dimensionality reduction and accurate classification in MTS analysis.
• To improve the performance of classifiers in diverse applications, including biomedical signal processing.

Main Methods:

• Random projection for dimensionality reduction in the time domain.
• Novel base classifiers in the frequency domain, including Whittle likelihood (WL), Kullback-Leibler (KL) divergence, eigen-distance (ED), and Chernoff (CH) divergence.
• Optimal weighted majority voting schemes for pooling classifier information in multiclass settings.

Main Results:

• The proposed methods demonstrate high efficacy in classifying high-dimensional MTS data.
• Successful application in both simulated binary/multiclass problems and real-world Electroencephalogram (EEG) data.
• Achieved accurate classification by combining dimension reduction with novel frequency-domain classifiers.

Conclusions:

• Random projection ensemble classifiers offer a powerful approach for high-dimensional MTS data.
• The novel frequency-domain base classifiers and voting schemes enhance classification accuracy.
• The method is effective and applicable to complex real-world datasets like EEG signals.