TXtreme: transformer-based extreme value prediction framework for time series forecasting
Abstract Time Series Forecasting (TSF) is crucial in various real-world applications such as climate forecasting and electricity demand prediction. Unlike traditional datasets, time series data points are influenced by their past values, necessitating specialized techniques to model these sequential...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-01-01
|
| Series: | Discover Applied Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s42452-025-06478-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Time Series Forecasting (TSF) is crucial in various real-world applications such as climate forecasting and electricity demand prediction. Unlike traditional datasets, time series data points are influenced by their past values, necessitating specialized techniques to model these sequential dependencies specifically addressing non-linear patterns, abrupt changes, and outliers. The latest advancements have significantly enhanced TSF using machine learning and other methods. However, forecasting extreme events remains challenging. Extreme values, although rare, have significant real-world impacts such as heavy rainfall, fluctuations in electricity demand, and traffic surges. This paper proposes a TXtreme framework that uses Long-Short memory network, feed-forward neural network, and transformer to improve time series forecasting under extreme values. The model also uses statistical methods to explain the distribution of time series values. Extensive experiments are conducted using datasets from different domains to show the robustness of the proposed methodology. Results, derived by testing TXtreme on five datasets of different domain, indicate that TXtreme significantly outperforms state-of-the-art methods in time series forecasting, with improvements of 5–25% in root means squared error or mean absolute error. The proposed framework enhances TSF capabilities and ensures better generalization ability in extreme event forecasting, potentially leading to improved decision-making in critical applications. |
|---|---|
| ISSN: | 3004-9261 |