Free tools Windows power users keep installed
One-click scans. No signup required.
Supervised machine learning trains a model on examples that pair input data with a known answer. The model learns a relationship between the inputs, called features, and the answer it is meant to predict, called the label. It can then use that relationship to predict labels for new examples. The two common task types are classification, which predicts categories, and regression, which predicts numeric values.
What are features and labels?
Features are the information supplied to a model; a label is the target answer associated with that information. Together, the features and label make a labeled example. For instance, a rainfall example might use temperature, humidity, air pressure, and wind as features, with the measured rainfall amount as its label. The model learns from many such pairs rather than being given a hand-written rule for every possible combination.
As an Amazon Associate I earn from qualifying purchases.
During training, the model makes predictions from the features and compares them with the known labels. The difference between a prediction and the correct label is called loss; training adjusts the model to reduce that difference. Google for Developers explains supervised learning, features, labels, and loss.
What kinds of predictions can supervised learning make?
The right task type depends on what the target label represents.
#1 Best Overall
Classification predicts a category
A classification model assigns an example to a category. An email spam filter, for example, predicts a label such as “spam” or “not spam.” Handwritten-digit recognition is another classification task: the model selects one digit category from a finite set. The scikit-learn 1.4.2 tutorial uses handwritten digits to illustrate classification.
Regression predicts a number
A regression model predicts a numeric value. If the target is rainfall amount, the model’s output is a number rather than a category. Predicting a house price is another example of a numeric target. The distinction is about the kind of answer being predicted, not which approach is inherently better.
Rank #2
- Use scikit-learn to track an example ML project end to end
- Explore several models, including support vector machines, decision trees, random forests, and ensemble methods
- Exploit unsupervised learning techniques such as dimensionality reduction, clustering, and anomaly detection
- Dive into neural net architectures, including convolutional nets, recurrent nets, generative adversarial networks, autoencoders, diffusion models, and transformers
- Use TensorFlow and Keras to build and train neural nets for computer vision, natural language processing, generative models, and deep reinforcement learning
How does training become a prediction?
Supervised learning usually involves three distinct stages. Training fits the model; evaluation checks its performance; inference uses the trained model to make predictions.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems- Train: Provide labeled examples. The model uses their features to predict the labels and adjusts its learned relationship based on the prediction errors.
- Evaluate: Compare predictions with known labels. A held-out test set—examples not used to fit the model—helps estimate performance on unseen examples. The scikit-learn tutorial describes splitting data into training and test sets.
- Infer: Give the trained model a new example’s features. It returns a predicted label, which may be a category or a number; the label for that new example is not yet known.
Evaluation is evidence about performance on the evaluated data, not a guarantee of equal performance in every real-world setting. Results depend in part on whether the examples represent the conditions in which the model will be used.
Rank #3
How is supervised learning different from unsupervised learning?
The key question is whether the training examples include the target the model is meant to predict.
| Approach | What the training or feedback provides | Typical aim |
|---|---|---|
| Supervised learning | Examples paired with known target labels | Predict a specified target, such as a category or numeric value |
| Unsupervised learning | Data without corresponding target values | Find groupings or other structure in the data |
| Reinforcement learning | Actions in an environment and rewards or penalties | Learn behavior through feedback from the environment |
These approaches address different kinds of problems; the distinction does not mean that one is universally better. Google’s overview of machine-learning types describes supervised, unsupervised, and reinforcement learning.
Rank #4
Why do data coverage and evaluation matter?
A large dataset does not necessarily cover the situations a model will encounter. For example, data collected across many years but only during one month may fail to represent other seasons. Diversity and coverage of relevant conditions matter alongside the number of examples.
Use held-out examples that reflect the intended use as part of evaluation, and treat the result as an estimate tied to those data—not proof that performance will transfer to every population or setting. There is no universal minimum number of training examples established here; what is sufficient depends on the task and the data.
Quick Recap
Best Value
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
