Machine Learning Algorithms

Data augmentation

Sometimes the original dataset has only a few non-linear features and it's quite difficult for a standard classifier to capture the dynamics. Moreover, forcing an algorithm on a complex dataset can result in overfitting the model because all the capacity is exhausted in trying to minimize the error considering only the training set, and without taking into account the generalization ability. For this reason, it's sometimes useful to enrich the dataset with derived features that are obtained through functions of the existing ones. PolynomialFeatures is an example of data augmentation that can really improve the performances of standard algorithms and avoid overfitting. In other cases, it can be useful to introduce trigonometric ...