Time Series Classification : Comparative Study of Some Matching Methods
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Faculty of Sciences
Abstract
Time series data, characterized by sequential measurements over time, are fundamental in various
domains such as finance, healthcare, and environmental monitoring. The increasing volume of
time series data, driven by IoT and sensor technologies, necessitates efficient and accurate analysis
methods. However, challenges such as high dimensionality, noise, temporal misalignments, and
varying lengths complicate classification tasks.
This thesis investigates prominent time series classification techniques, evaluating methods
including Euclidean distance, Dynamic Time Warping (DTW), Longest Common Subsequence
(LCSS), and the Shape Exchange Algorithm (SEA) using the 1-Nearest Neighbor (1NN) classifier
on 85 datasets from the UCR Archive.
We propose two novel variants, PAA-SEA and RDP-SEA, though experiments confirm the superiority
of the original SEA. Additionally, we explore a hybrid DTW-LCSS approach, which outperforms
standalone methods.
We also assess the previously untested MSR (Mean Squared Residue) method in this field,
demonstrating competitive accuracy.