Supervised machine learning

Try Domo for yourself.

Completely free.

What is supervised machine learning?

Supervised machine learning is the most common type of machine learning. It utilizes structured data with a set of input variables to predict the value of an output variable. This kind of machine learning can often be grouped into regression and classification. Regression is when the variable to predict is numerical or of a real value like dollars or weight. Classification is when the variable to predict is categorical, like green or yellow. Common types of problems that can be built with regression and classification include recommendation and time series prediction. In addition to regression and classification is forecasting. Forecasting is the process of making future predictions based on past and present data, and it is typically used to analyze trends.

Supervised learning, also known as inductive learning, assists with real-world applications, including credit risk assessment, disease diagnosis, face recognition, and automatic steering.

Supervised machine learning vs. Unsupervised machine learning

Unsupervised machine learning differs from supervised machine learning in that its purpose is to discover hidden patterns and identify like groups within unstructured data. With supervised learning, data is labeled, and algorithms learn to predict an output from the input data. With unsupervised learning, data is unlabeled, and algorithms learn to find structure within the input data since there are no corresponding output variables. 

Unsupervised learning can be grouped into clustering and association. Clustering is discovering inherent groupings in the data, like grouping customers by shopping preferences. Association is discovering rules to describe large data quantities like customers that purchase item A may also purchase item B.

Supervised learning algorithms

Within supervised machine learning, different categories of algorithms are used to make specific calculations and solve problems. They include: 

Linear models

    Linear regression is a linear model that works to find the best fit line through data points to generate numerical predictions. For this reason, linear regression only works for regression problems.

  • Logistic regression is a linear model that is an adaptation of linear regression. It works for categorical problems and generates an S-shaped line of best fit to predict the likelihood of a given data point belonging to one category. For this reason, logistic regression only works for classification problems.

Tree-based models

  • Decision trees are tree-based models, and they are a series of if-then rules based on features. In other words, a tree is formed to match all possible outcomes of a decision. Decision trees work for both regression and classification problems.
  • Random forest is a tree-based model and tree-based ensemble algorithm that uses random subsamples of features to determine the most significant feature to split on. Random forest works for both regression and classification.
  • Gradient boosting is a tree-based model and another tree-based ensemble algorithm that builds one-node trees sequentially. This means that each tree learns from the incorrect predictions of the previous tree. Gradient boosting works for both regression and classification.

Artificial neural networks

  • Artificial neural networks are a form of deep learning, and these networks can pick up on complex relationships. Artificial neural networks are mainly used for image classification and speech recognition, and they work for both regression and classification. 



How supervised learning algorithm work

Linear models are more straightforward models, but they have a challenging time capturing complex relationships. Both linear regression and logistic regression are too simplified and don’t do well with correlated features. 

Tree-based models often make strong predictions. For example, a random forest can make strong predictions and is also fast to train, but the interpretation can be somewhat unclear. Gradient boosting can also make strong predictions, but it is slow to train, and the interpretation can also be somewhat unclear. When it comes to decision trees, they are easy to understand but are often prone to overfitting, and the model won’t generalize well for new data. There is also an ensemble model for decision trees wherein many trees can be built to combine predictions, and this can be done for both regression and classification.

Artificial neural networks can handle complex problems, but they are slow to train and extremely difficult, if at all possible, to interpret.

Types of Supervised Learning Use Cases

As the most common type of machine learning, there are more examples of supervised machine learning use cases. Several different industries utilize supervised learning to assist with the following:

  • Forecasting sales
  • Forecasting supply and demand
  • Assessing and detecting fraud
  • Minimizing customer churn



Unsupervised learning algorithms

Unlike supervised machine learning, there aren’t many different categories of algorithms within unsupervised machine learning. The common type of algorithm used in unsupervised learning is K-Means or clustering. Clustering separates data samples into several different clusters, which are characterized by centers or centroids.

Types of unsupervised learning use cases

Unsupervised learning primarily focuses on the data extracted from images, audio, and video. Examples of unsupervised learning, or clustering, being used include:

  • Customer segmentation
  • Recommendation systems
  • Anomaly or fraud detection

What is reinforcement machine learning?

Reinforcement learning can often be classified as the most ambitious type of learning in that it rewards from a sequence of actions. Reinforcement learning lets machines automatically determine ideal behaviors within a specific context to maximize the machine’s performance.

Reinforcement learning is also powered by supervised models. In other words, it is an incremental add-on to supervised learning models. Reinforcement learning adds the ability to influence a given model or algorithm based on the actions of the primary supervised model.

What does supervised machine learning look like in practice?

Supervised machine learning helps businesses collect data, produce various data outputs, optimize performance criteria, and solve different computation problems. Therefore, when choosing a supervised learning algorithm, it’s important to know not only the problem type you’re solving for but also the level of accuracy, interpretability, and training time you need.

Domo’s comprehensive machine learning solutions bring in data, prepare it, and build models and solutions based on model pipelines that are useful and practical. Domo has a Machine Learning playbook that supports anyone in an organization with preparing data, running a model in a ready-made environment, and visualizing it back within Domo. Additionally, there is AutoML which helps with pre-processing data, choosing a model, and hyperparameter tuning — everything you need to help you move from data to models to outcomes even faster.



Level Up your Analytics Strategy with Augmented BI


Domo Ranked #1 Vendor in the 2022 Dresner Advisory Services’ Self-Service Business Intelligence (BI) Market Study


Are you ready for data science?


Closing the Data Decision Gap

Ready to get started?
Try Domo now or watch a demo.