Effective Amazon Machine Learning

Implementing recursive feature selection with Boto3

In many real-world cases, the original dataset could have a very large number of variables. As dimensionality increases, so does the need for a larger sample set. This is called the curse of dimensionality, a classic predictive analytics problem. Simply put, if there is not enough diversity to infer a representative distribution for some variables, the algorithm will be unable to extract relevant information from the said variables. These low-signal variables drag down the algorithm's performance without adding any data fuel by adding useless complexity to the model. One strategy is to reduce the number of variables on which to train the model. However, that implies identifying which features ...