Automated Machine Learning
Feature Selection
Use only relevant features in your models.
Drop noisy features from the analysis
Use only relevant features in your models.
Drop noisy features from the analysis
The feature selection is a process of selecting only relevant features (with signal) for the ML model construction.
The AutoML feature selection works procedure in two steps. First step is to train ML model with random feature inserted to the data.
The permutation-based features importance is computed for a such model.
The features with importance smaller than random feature are removed. The ML models are trained only on selected features.
Number of selected features are determined automatically.
Convert categorical features into numeric with label encoder, one-hot encoder or target encoder. Proper encoder type is automatically selected based on feature cardinality and AutoML training stage.
Convert date and time features into numbers understood by ML algorithms. Get features link year, month, day, weekday, day of the year, hour, and difference to the earliest date automatically extracted from your data.