Feature
One input column a model reads — a single measured or constructed number describing each row.
distance_km, is_weekend, word_count. Statisticians call these predictors or independent variables; in machine learning they're features, and the words mean the same thing.
The important property is that a feature is something you chose. A model can only combine the columns in front of it, so a pattern that isn't expressed in any feature cannot be learned — which is why constructing a good column often beats swapping the algorithm.
Worth separating from the raw data it came from. A timestamp sitting in your database is not a feature until someone decides what to extract from it, and hour_of_day and days_since_signup are two different features from the same source column.