It is usually the data type of a problem that determines the suitable technology (e.g., vector data, list data and etc.)
Feature Construction is one of the key steps in data analysis process, largely conditioning the success of any machine learning endeavor.
One should beware of NOT losing information at the feature construction stage. It may be a good idea to add the RAW FEATURES to the preprocessed data - AND use FEATURE SELECTION (or REGULARIZATION) in the machine stage, This is simplely because ADDING all those features comes at a price - it increases the dimensionality of the patterns and thereby immerses the relevant information into a sea of possibly irrelevant, noisy or redundant features - and thus increase the difficulty for machine learning models to search for the optimal solution (in a bigger search space).
data value type can be binary, categorical or continous
To understand the organization of the tutorial, there are four aspects of feature extraction:
filters: methods that select features without optimizing the performance of a predictor.
wrappers utilize a learning machine as a blackbox to score subsets of features according to their predictive power
embedded methods perform feature selection in the process of training and are usually specific to given learning machines
ensemble of wrappers/embedded methods : wrappers and embedded methods may be yield very different feature subsets under small perturbatioins of the dataset. One way of minimizing the effect is to use ensemble methods
Some basic techniques and discussions about how feature selection can be done and why it is this way
Individual Relevance Ranking
Pearson correlation coefficient
Fisher-coefficient
PCA
can be used to perform such linear transformationMultivariate Feature Ranking
Eliminating meaningless features is note critical
A filter as simple as the Pearson correlation coefficient proves to be very effective