Machine Learning Nowcasts Visibility in Six South Korean Cities
A study from arXiv introduces a machine learning framework to nowcast atmospheric visibility in six major South Korean cities, addressing challenges of class imbalance and distribution shift. Using 2018-2020 training data, researchers applied SMOTENC and CTGAN to handle rare low-visibility events. An ensemble of machine learning and deep learning models was tested on 2021 data, revealing a marked decline in performance due to distributional shift between training and testing periods.
Key facts
- Visibility nowcasting framework developed for six major South Korean cities
- Training data from 2018-2020 used
- SMOTENC and CTGAN applied to handle class imbalance
- Ensemble of machine learning and deep learning models employed
- Tested on 2021 dataset
- Predictive performance declined in test set compared to cross-validation
- Degradation attributed to distributional shift between training and testing periods
- Study published on arXiv (2605.21507)
Entities
Institutions
- arXiv
Locations
- South Korea