Abstract

The Wavelet Entropy Clustering (WEC) technique enhances the pattern recognition (PR) methods in science and engineering domains. WEC helps identify hidden patterns in time series (TS) data. State-of-the-art WEC methods rely on advanced algorithms and wavelet function selection. Selecting pre-defined wavelets for PR might not capture all the patterns in the TS data. This limitation affects the accuracy of the TS data analysis. Existing data-driven-based PR methods handle complex TS data. They construct wavelets tailored to the data. This approach improves the accuracy of PR in complex TS datasets. Balancing all the desired wavelet properties like orthogonality and compact support becomes challenging. This inherent trade-off in achieving all properties limits the effectiveness of data-driven methods. Wavelet set theory (WST) offers a framework for addressing these challenges. WST tailors wavelets from a TS signal. Tailor wavelets perform better signal decomposition, capture complex features, and conduct multi-scale analysis. In this work, we propose WaveClust, a new approach which combines WST into the WEC framework. WaveClust extracts hidden features from TS data using multi-level wavelet packet entropy. We compare the performance of WaveClust with existing wavelet-based methods. WaveClust outperforms conventional approaches in PR tasks. It improves the ability of PR tasks to identify complex patterns in TS data. WaveClust aids researchers in analyzing complex TS data for scientific and engineering applications.