Wzmacnianie zdolności predykcyjnych modeli dyskryminacyjnych przez wyodrębnianie zmiennych objaśniających z sieci neuronowych
Improving classification accuracy through neural networks’ feature extraction

Summary/Abstract: Combining different methods has proven to be a good idea that yields models with better prediction abilities. By deleting the output layer we can use neural networks as a method for feature extraction designed to work well for classification problems. By doing this we obtain dimensionality reduction similar to PCA, but with the new features being built with the specific purpose – for classification task. We can expect this technique to generate features with high discrimination power. The main goal of the research is to analyze whether this neural networks’ feature extraction method can significantly improve the classification accuracy. The results show that it does the job sometimes, but comes with no warranty. Still it can be treated as an interesting, nonlinear alternative to PCA and a vaulable data preprocessing (dimensionality reduction) technique.

• Details
• Contents