feature engineering
Deriving predictive attributes manually
Feature engineering is the process of selecting, transforming, or creating input variables to improve a machine learning model's performance.
Raw data rarely arrives in ideal form—a date might become "day of week" or "days until holiday," an address might become "distance from city center." Good feature engineering requires domain knowledge and often matters more than algorithm choice: We spent weeks on feature engineering before touching the model.
- preprocessingInitial data preparation steps
- data wranglingTransforming raw information for analysis
- transformationThorough change in form
- variablesChangeable elements in equations
- attributesInherent characteristics or properties
normalization · encoding · scaling · binning · imputation · polynomials · interactions · selection · PCA · vectorization · tokenization · aggregation · pipeline · pandas · columns · categorical · numerical · crafting
- data sciencealgorithm, data analytics, data, machine learning, dataset
- AI conceptsneural network, artificial general intelligence, neural net, superintelligence, CNN
- ML modelsneural network, neural networks, decision tree, decision trees, linear regression
- ML settingshyperparameters, learning rate, epochs, dropout, regularization
- Outlier Detectionz-score, MAD, box plot, leverage, boxplot
- tree ensemblesdecision tree, decision trees, ensemble, bagging, boosting
feature, features, featured, featuring, featureless, featurette, engineer, engineers, engineered