Knowledge-Constrained-Data-Augmentation

Data and codes for the Paper "Enhancing the prediction accuracy of concrete properties with knowledge constrained data augmentation and tabular foundation model"

1. The folder datasets stores the 15 datasets used in the study.
2. The data_aug.py defines four data augmentation models: GaussianCopula, CTGAN, CopulaGAN, and TVAE.
3. The filter.py defines three anomaly detection methods: RANSAC, IF, and LOF.
4. The model.py defines the TabPFN framework
5. The benchmark.py defines eight benchmark ML models, which are used to compare with TabPFN
6. The prediction.py defines the whole experimental process.
7. All synthetic datasets can be downloaded here. The password is MT1G.