MMOOZZGG.com

Glossary

What are features in machine learning

In short

Features in machine learning are the properties of an object, written as numbers or categories, that a model uses to make a prediction: a customer’s age, an order amount, the day of the week.

Machine learning · features

MMOOZZGG.com editorsChecked: 2 min read

For a child

Imagine the game “Guess the animal”. You hear: it has a tail, it meows and it loves milk. You answer at once: a cat. The tail, the meowing and the milk helped you guess. A computer also guesses from such hints, but people write them into a table as numbers and words for it. These hints are called features.

In plain words

For a model to predict anything, each case must be described with data it understands. For example, an apartment is described by its area, floor, district and year of construction. Each such property is a feature. The model looks at thousands of apartments with known prices and finds how these properties relate to the price. Then it uses the same properties of a new apartment to estimate its value. Useful properties make the forecast more accurate, and random ones only get in the way. In statistics, features are also called predictors.

For business

The quality of a company’s prediction model often depends on its features, not on the choice of algorithm. A churn model will not see the risk if the data has no purchase frequency, support requests or date of the last order. So the main work in such projects is on the company’s side: find where this data lives, in the CRM, the accounting system or the point-of-sale system, and gather it into one table. This shapes the budget too: integrations and data cleaning often cost more than training the model. A separate risk is information known only after the fact, such as a “returned the item” status in a returns forecast. With it, the model shines on history and is useless in real life. Teams look for such traps before launch, when they plan predictive analytics.

For a specialist

A feature is a function that gives each object in a sample a numeric or categorical value; in the data matrix, features are the columns and objects are the rows. The scikit-learn glossary defines them this way and lists synonyms: attributes, predictors, regressors and independent variables. The types are numeric, categorical (encoded by one-hot or target encoding), ordinal, binary, temporal, text and embeddings. Feature engineering creates new columns from raw data: aggregates, ratios, lags and frequencies. Feature selection methods are filters, wrappers and embedded methods. Guyon and Elisseeff (JMLR, 2003) gave an overview and named three goals: predict more accurately, compute faster and cheaper, and better understand the process that produced the data. Deep networks learn representations by themselves (representation learning), but on tabular tasks careful manual data preparation still decides a lot. A typical preparation error is data leakage, when an input column gets information that is not available at the moment of the forecast.

Do not confuse. Do not confuse this with features in the everyday sense, such as product features or signs of an illness: here they are the data columns a model learns from.

Example

A chain of clinics predicts missed appointments. The model’s features are the day of the week, the booking time, the days between booking and visit, the patient’s past no-shows and whether a reminder was sent. With them, the model marks risky bookings in advance, and an administrator calls those patients.

Related terms

Questions and answers

What are features in machine learning? Features are the model’s input data: the properties of each object, written as numbers or categories. For a customer, they are age, city, number of purchases and average order value.
Are features the same as attributes or predictors? Yes, attributes and predictors are other names for features, used in databases and statistics. Analysts say they “build a feature” when they invent and compute a new property of an object from raw data.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us