Using Deep Learning Models for Structural Break Detection in Time Series

Date

2024-12-06

Author

Yazıcı, Pınar Cemre

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

41
views

24
downloads

Cite This

This study addresses the detection and localization of structural breaks in time series data through a comprehensive pipeline, encompassing synthetic data generation, model training, and validation. Synthetic datasets were created to emulate diverse characteristics such as trends, anomalies, and structural shifts, including both gradual and sudden changes. Neural network architectures, including convolutional and recurrent models, were trained using Python to identify structural breaks, while statistical tests were implemented using R for comparison. The models’ performance was evaluated on detection metrics (precision, recall, F1-score) and locational accuracy. The results highlight the proposed models’ robustness in detecting structural breaks across varying time series lengths, outperforming traditional statistical tests in most metrics. This research contributes to advancing break detection methodologies, offering a scalable and accurate alternative to conventional approaches.

Subject Keywords

time series, structural break detection, change points, deep learning, neural networks, lstm

URI

https://hdl.handle.net/11511/112962

Collections

Graduate School of Informatics, Thesis