Supervised learning is a method of building a model that predicts outputs from inputs, using input data together with the corresponding correct outputs.

A trained model is given new inputs and is used to predict outputs that were not provided during training.

To make this work in practice, it is not enough to choose and train a model. A complete workflow is needed: defining what to predict, preparing the data, and evaluating the resulting model.

2.1.1 Components of Supervised Learning

The basic structure of supervised learning consists of input data, ground-truth data, a model, a learning method, and an evaluation method.

Input data is the information given to the model as the basis for prediction, and ground-truth data indicates the value that should be predicted for each input.

Ground-truth data, which represents what is to be predicted, is also called the target variable or simply the target.

The model represents the relationship that converts inputs into predictions. In learning, the model’s parameters are adjusted based on the discrepancy between the model’s predictions and the ground truth.

In evaluation, we check how well the model predicts on data that was not used for training.

2.1.2 Problem Definition and Data Preparation

The starting point of supervised learning is defining the scope of the problem to be solved.

This involves clarifying what is to be predicted, what information can be used as input at prediction time, and what level of performance is required.

Once the problem is defined, data in which inputs are paired with ground-truth values is collected.

Collected data may contain missing values, duplicates, inconsistent notation or formats, outliers, and other issues. Therefore, rather than using the data as is, its contents are checked, the necessary processing is performed, and the data is put into a form the model can handle.

Collecting, loading, organizing, and transforming data in this way are important steps that support the supervised learning workflow.

2.1.3 Building and Training the Model

A model suited to the prediction target and the characteristics of the data is chosen for the prepared data.

In supervised learning, the model’s prediction obtained from an input is compared with the corresponding ground truth.

The difference between the two is quantified by a loss function, and the learning algorithm adjusts the model’s parameters so that the loss becomes smaller.

Training is therefore not simply a matter of feeding data into a model. It is a process of fitting the relationship between inputs and outputs to the data by repeating prediction, comparison with the ground truth, and parameter adjustment.

The results of training depend not only on the data used, but also on how the data is transformed and on the choice of model type, loss function, and learning method.

2.1.4 Evaluation and Use of the Trained Model

Even if a model predicts well on the training data, it may not work as well on new inputs.

For this reason, during model development, predictive performance is evaluated using data that was not used for training.

Depending on the purpose and the characteristics of the data, multiple models or methods may also be compared.

Once the required performance has been confirmed, the trained model is used to predict outputs from new inputs. At this stage, no ground-truth data is input; instead, new input data is given to the model whose parameters were determined by training, and a prediction is obtained.

Summary

In supervised learning, a model that predicts outputs from inputs is trained using the correspondence between input data and ground-truth data.

This basic structure is not complete with inputs and ground truth alone. It is necessary to define the problem to be predicted, collect and organize the data, choose a model and learning method, and evaluate performance separately from training.

By viewing supervised learning not as a single algorithm but as a workflow that runs from problem definition to the use of the trained model, the role of each step can be clearly understood.

About This Article

References

• Styliani I. Kampezidou et al., “Fundamental Components and Principles of Supervised Machine Learning Workflows with Numerical and Categorical Data”
• Lorenzo Ciampiconi et al., “A Survey and Taxonomy of Loss Functions in Machine Learning”

(Last accessed: October 16, 2026)


*This article was compiled based on the author’s understanding, with reference to the literature listed above.