AI Terms

Data a person intentionally and proactively shares with an organization. This may include things like preference center data, what the person intends to purchase, and how they want the brand to recognize them and interact with them.

A way of indicating how much a particular feature contributes to a model.

Also called a data field, feature, or variable, this is a specific category of information found in a record. For instance, if you have a record of a donor that includes their name, height, weight, address, and phone number, those five pieces of information are variables. We prefer the term data field.

Data used for tuning hyperparameters and comparing different models to determine the best ones. This has to be different from training data, and can’t be used in the training phase, or you’re at risk of overfitting (creating a model that matches training data so closely that the model is unable to make correct predictions with…

Where a program, given a data set, can automatically find patterns and relationships in that data set.

Data used by a machine learning algorithm to train a model. It consists of positive training data and negative training data, and is used to teach the model patterns and to determine which features are more important during prediction.

Aggregated information collected from multiple sources by organizations that do not have a direct relationship with the person whose data’s being collected. This differs from first-party data (information an organization has collected about their contacts themselves) or second-party data (collected by another organization.)

Also called a hold-out, it’s a third and final test set used to simulate the model’s behavior on completely unseen data. (Data that wasn’t used in building the model or in choosing which model to use.)

The label you want to predict using a machine learning model.

The data used to generate recommendations using a machine learning model.