Feature extraction is an extension to feature selection. Simply speaking, feature extraction tries to project the original dataset onto a lower-dimension space such that some objective function is optimized. Therefore we need to employ some optimization method to find such a lower-dimension basis for projection.
Since the objective function used in feature extraction depends on the types of datasets, we usually classify feature extraction methods into two types:
Details of PCA and LDA will be covered in the subsequent sections.
- For datasets with no labels (such as datasets for clustering), we usually apply principal component analysis (PCA) for feature extraction.
- For datasets with desired class labels (such as datasets for classification), we usually apply linear discriminant analysis (LDA) for feature extraction.
Data Clustering and Pattern Recognition (資料分群與樣式辨認)