Feature Engineering
Feature Engineering is the art of creating features from raw data, so the predictive models can deeply understand the dataset and perform well on unseen data.
Types in Feature Engineering
Feature Extraction
For example the dimensionality reduction.
Feature Selection
Feature Construction
Feature Scoring Functions
- F-score
- mutual information score
- Chi-square score
Tools
Deep Feature Synthesis
DFS is an automated feature engineering mothod that generates meaningful features from structured data by leveraging relationships within the data, such as joins or temporal dependencies.
DFS vs DL
DFS explicitly builds human-readable features from structured data using predefined rules, requiring no training, and is lightweight and interpretable. DL implicitly learns abstract features through neural networks, excels with unstructured data and relies on iterative trainingk offering high performance but low interpretability.