A Fast Framework for Abrupt Change Detection Based on Binary Search Trees and Kolmogorov Statistic.

Area of Science:

• Data Mining
• Statistics
• Bioinformatics

Background:

• Change-point detection is crucial for analyzing large-scale bioelectric signals.
• Existing methods like Kolmogorov-Smirnov (KS) are often time-consuming for large datasets.

Purpose of the Study:

• To propose a fast and efficient framework for abrupt change detection in bioelectric signals.
• To improve upon the speed and accuracy of current change-point detection methods.

Main Methods:

• Developed a novel framework named BSTKS (binary search trees and Kolmogorov statistic).
• Utilized binary search trees (BSTs) and a modified KS statistic.
• Employed multilevel Haar Wavelet Transform (HWT) for BST construction.
• Introduced search criteria based on statistic and variance fluctuations.

Main Results:

• BSTKS demonstrated significantly faster computation times compared to KS, t-statistic (t), and Singular-Spectrum Analyses (SSA).
• Achieved higher hit rates, lower error rates, and superior accuracy in detecting abrupt changes.
• Validated performance on synthetic time series and real electroencephalograph (EEG) data.

Conclusions:

• The proposed BSTKS framework provides a rapid and effective solution for abrupt change detection in bioelectric time series.
• BSTKS is highly beneficial for information inspection in various bioelectric signal analyses.
• This method enhances the efficiency of analyzing large-scale bioelectric datasets.