Supervised learning
Supervised learning is a machine learning method in which a model learns from labelled training data. Each input has a known, correct output assigned to it. The model recognises the relationships between input and output and can then make predictions for new, unknown data. Typical applications are classification and regression.
Also known as: supervised learning, supervised ML
How does supervised learning work?
In supervised learning an algorithm is presented with a dataset in which every example contains both the input features and the corresponding correct answer. That answer is called the label or target value. An example: to train a model to detect spam, you need numerous emails already marked by hand as spam or not spam.
During training the model compares its own predictions with the actual labels and adjusts its internal parameters step by step to minimise the error. This process is repeated until the model gives reliable predictions. Its performance can then be checked on a separate test dataset it has not seen before.
Classification and regression compared
Supervised learning splits into two major task types. In classification the model assigns an input to one of several categories, such as classing an email as spam or recognising an object in an image. The output here is a discrete value, that is a class.
In regression, by contrast, the model predicts a continuous numerical value. Examples are predicting a property price from location and size or forecasting energy consumption. Both methods rest on the same principle of labelled data but differ in the kind of target.
Which algorithms are used?
A great many proven algorithms exist for supervised learning. These include linear and logistic regression, decision trees, random forests, support vector machines and artificial neural networks. Choosing the right method depends on the amount of data, the problem's complexity and the demands on accuracy and traceability.
Neural networks and deep learning architectures in particular underpin many modern AI applications today. They are trained under supervision, for instance for image or speech recognition, and need large amounts of carefully labelled data for it.
Challenges and data quality
The biggest hurdle in supervised learning is obtaining high-quality training data. Labelling by hand is time-consuming and expensive, especially with large volumes or complex tasks. Faulty or biased labels inevitably lead to worse models, since the algorithm learns the wrong patterns too.
Another risk is overfitting. The model learns the training data by heart too precisely and loses the ability to generalise to new data. Thoughtful data preparation and validation are therefore decisive for success.
Practical use and advice from Elisabit
Supervised learning is in countless applications today, from fraud detection through medical diagnostics to image classification in online shops. For companies it offers a comparatively accessible entry into AI, provided the labelled data needed exists or can be produced.
At Elisabit we help companies pick the right learning method for their use cases, prepare data sensibly and AI solutions in practice. That way supervised learning can be built deliberately into existing business processes and put to work.
Frequently asked questions
What is the difference between supervised and unsupervised learning?
In supervised learning the model learns from labelled data with known answers. In Unsupervised learning there are no given answers, and the model finds structures in the data on its own. Supervised learning suits prediction, unsupervised learning pattern recognition.
What data does supervised learning need?
You need a dataset in which every input carries the correct output, the so-called label. Depending on the task that can be annotated images, classified text or tabular data with target values. The quality and quantity of that labelled data largely determine the model's accuracy.
What is supervised learning used for in practice?
Supervised learning is found in spam detection, image and speech recognition, credit scoring and medical diagnostics. It is used wherever reliable predictions are to be derived from historical example data.
What does overfitting mean in supervised learning?
Overfitting is a model adapting too closely to the training data. It then learns the examples by heart instead of recognising general patterns and gives poor results on new data. Methods such as cross-validation and regularisation contain the problem.
Related terms
Machine learning enables systems to learn from data and make predictions without being explicitly programmed.
Unsupervised learning finds structures and patterns in data on its own, without given answers.
Reinforcement learning lets an agent learn optimal behaviour through trial, error and reward.
Deep learning uses deep neural networks to recognise complex patterns in large volumes of data automatically.
A neural network is a computational model of connected neurons, modelled on the brain, that learns from data.
Put AI to work for your business?
We help you integrate artificial intelligence into your processes, your marketing and your website — strategically and securely.
Your contact
Stefan
I look forward to hearing about your project and finding the best solution together.
