{"id":21235,"date":"2020-01-30T16:00:55","date_gmt":"2020-01-30T16:00:55","guid":{"rendered":"https:\/\/www.arimetrics.com\/glosario-digital\/predictive-model"},"modified":"2026-09-29T14:20:22","modified_gmt":"2026-09-29T14:20:22","slug":"predictive-model","status":"publish","type":"encyclopedia","link":"https:\/\/www.arimetrics.com\/en\/digital-glossary\/predictive-model","title":{"rendered":"Predictive Model"},"content":{"rendered":"

\"PredictiveDefinition:<\/strong><\/p>\n

A predictive model<\/strong> is a mathematical or computational representation that learns relationships from data to estimate an unknown outcome. It can calculate a value, assign a category, rank cases, or estimate a probability.<\/p>\n

The outcome may concern the future, such as demand next month, or a variable that already exists but is not yet known, such as whether a transaction is fraudulent. A prediction<\/strong> is not a certainty: it depends on the data, method, and conditions in which the model is used.<\/p>\n

The model is one component of predictive analytics<\/a><\/strong>. Analytics also includes problem definition, data preparation, evaluation, interpretation, and monitoring of results.<\/p>\n\n

Model operation<\/h2>\n

Development begins with a question that can be translated into a target variable. The target<\/strong> determines the appropriate data, method, and metric; applying an algorithm to a collection of records is not enough.<\/p>\n

    \n
  1. Define the outcome:<\/strong> specify what will be estimated, its unit or class, and the decision for which it will be used.<\/li>\n
  2. Prepare the data:<\/strong> select observations and features, handle missing values, encode categories, and prevent information unavailable at prediction time from entering the model.<\/li>\n
  3. Train the model:<\/strong> fit its parameters on part of the data using statistics or machine learning<\/a>.<\/li>\n
  4. External evaluation:<\/strong> measure performance on data that was not used to fit the model.<\/li>\n
  5. Deploy and monitor:<\/strong> integrate the prediction into a process and check whether performance changes with new data.<\/li>\n<\/ol>\n

    A separate evaluation<\/strong> helps detect overfitting. Cross-validation<\/a> repeats that comparison across different splits, although it must account for temporal or grouped data structures.<\/p>\n

    Types of predictive models<\/h2>\n

    Models differ according to the outcome they produce. Four families<\/strong> cover many common applications:<\/p>\n