Feature engineering
Transforming raw data into input variables, called features, that help a model learn useful patterns. Examples include days since last purchase, rolling averages or ratios between related measures.
Why it matters
Well-chosen features often matter more to model performance than the choice of algorithm, especially with structured business data. They are also where domain knowledge enters a model.
Features must be computed the same way in training and in production, and only from information available at the moment of prediction. Leaking future information into training is a common reason models perform well in testing and poorly in use.
In practice
For example, a UK subscription business predicting cancellations might create features such as the change in weekly usage over the last month and the number of support contacts in the past fortnight, calculated only from data available on the prediction date.
Where Rodan fits
Rodan builds reusable, tested feature pipelines as part of Data and Analytics Engineering and model delivery.