It involves training a model using labeled data that has clearly defined inputs and outputs, where the input variables are mapped to an output variable. It is often used in classification and regression problems where the model learns to predict the output variable based on input variables.