This article discusses the efficacy of a novel data augmentation technique using PV-sets derived from various decomposition methods. The results showed notable improvements in model performance through this augmentation, particularly when proper optimization of artificial neural networks (ANN) learning parameters is executed. While the method generally enhances predictive models, it’s crucial to understand that it may not be beneficial for all algorithm types, such as random forest, where resilience to overfitting limits the advantages of increased data size. Recommendations for cross-validation and latent variable selection are provided for optimal results.
The experimental results confirm the benefits of PV-set augmentation, although optimization of ANN learning parameters is crucial to ensure significant advantages in model performance.
It's important to note that PV-set augmentation does not always improve performance, particularly in methods robust against overfitting, like random forest.
Collection
[
|
...
]