In this chapter, we have discussed data preprocessing, or the art of delivering the most useful possible data to our machine learning algorithms. We discussed the importance of appropriate feature selection and the relevance of feature selection, both to overfitting and to the curse of dimensionality. We looked at correlation coefficients as a technique to help us determine the appropriate features to select, and also discussed more sophisticated wrapper methods for feature selection, such as using a genetic algorithm to determine the optimal set of features to choose. We then discussed the more advanced topic of feature extraction, which is a category of algorithms that can be used to combine multiple features into new individual features, further reducing the dimensionality of the data.
We then looked at some common scenarios you might face when dealing with real-world...