Transformer-based Short-term Drought Prediction with Feature Selection 


Vol. 14,  No. 4, pp. 282-288, Apr.  2025
https://doi.org/10.3745/TKIPS.2025.14.4.282


PDF
  Abstract

There have been many studies on short-term drought prediction. In the past, research mainly focused on numerical modeling. However, due to the limitation of prediction accuracy, studies with artificial intelligence technology have recently gained traction. In the beginning, machine learning models using a single algorithm and multilayer perceptron-based drought prediction research were conducted, and then ensemble machine learning techniques and deep learning algorithms were introduced. Recently, transformer-based deep learning algorithms, which have demonstrated outstanding performance, have garnered significant attention. In this study, longitudinal weather observation data and standard precipitation data provided by the Meteorological Administration from 2015 to 2023 were preprocessed and utilized. During the data pre-processing phrase, methods of processing missing values and outliers were implemented, new features are generated from existing features, and feature extraction based on the wrapper method using a genetic algorithm and XGBoost is performed. The key features were selected based on therir F1-score as the evaluation metic and were then used for machine learning and deep learning experiments. The experimental results indicated that the transformer-based TFT deep learning algorithm, trained with the selected features, achieved the highest F1-score of 0.88.

  Statistics
  Cite this article

[IEEE Style]

S. J. Hyun, "Transformer-based Short-term Drought Prediction with Feature Selection," The Transactions of the Korea Information Processing Society, vol. 14, no. 4, pp. 282-288, 2025. DOI: https://doi.org/10.3745/TKIPS.2025.14.4.282.

[ACM Style]

Seo Jae Hyun. 2025. Transformer-based Short-term Drought Prediction with Feature Selection. The Transactions of the Korea Information Processing Society, 14, 4, (2025), 282-288. DOI: https://doi.org/10.3745/TKIPS.2025.14.4.282.

Forms

Search
(IN TITLE, AUTHOR, ABSTRACT,KEYWORDS)

Advanced Search ▼

POPULAR KEYWORDS
(TOP 10 KEYWORDS)

Recent Publications
(LAST 3 YEARS)

Old Journals

Indexing

All publications of TKIPS is indexed in DOI, EBSCO, Google Scholar, Crossref, and CrossCheck.

TKIPS is also selected as the Journal for Accreditation by NRF (National Research Foundation of Korea).

Related Journals