Awesome Feature Engineering

Overview

Awesome Feature Engineering provides resources for creating better features from raw data, a critical step that often determines machine learning model success.

What is Feature Engineering?

The process of using domain knowledge to extract features from raw data via data mining techniques to improve machine learning algorithm performance.

Core Techniques

Numerical Features

• Scaling and normalization
• Binning/discretization
• Polynomial features
• Log transformations
• Power transformations

Categorical Features

• One-hot encoding
• Label encoding
• Target encoding
• Frequency encoding
• Binary encoding

Datetime Features

• Extracting components (year, month, day)
• Cyclical encoding (sin/cos)
• Time since events
• Holiday indicators

Text Features

• TF-IDF
• Word embeddings
• N-grams
• Count vectorization
• Sentiment scores

Image Features

• Color histograms
• Edge detection
• SIFT/SURF features
• Deep learning embeddings

Advanced Methods

Automated Feature Engineering

• Featuretools
• tsfresh (time series)
• AutoFeat
• Feature-engine

Feature Selection

• Filter methods (correlation, mutual information)
• Wrapper methods (RFE)
• Embedded methods (L1 regularization)
• Importance-based selection

Feature Interactions

• Polynomial interactions
• Ratio features
• Difference features
• Product features

Domain-Specific

Time Series

• Lag features
• Rolling statistics
• Seasonal decomposition
• Fourier features

Geospatial

• Distance calculations
• Spatial aggregations
• Coordinate transformations
• Clustering-based features

Tools & Libraries

• scikit-learn preprocessing
• pandas for data manipulation
• Featuretools for automated generation
• category_encoders for categorical encoding
• tsfresh for time series

Best Practices

• Understand your data first
• Domain knowledge is crucial
• Avoid data leakage
• Cross-validate feature engineering
• Document transformations
• Monitor feature drift

Evaluation

• Feature importance analysis
• A/B testing with/without features
• Learning curves
• Validation set performance