When it comes to machine learning, you need to consider and understand the differences between the two main methods used: supervised and unsupervised machine learning.

Related course:

Supervised vs Unsupervised Learning

The core distinction between the two types is the fact that supervised learning is done by using a ground truth or simply put: there exists prior knowledge of what the output values for the samples should be. Thus, the purpose of supervised machine learning is to discover a function that, given a sample data and a couple of outputs, properly approximates the link between the input and the output data.

On the other hand, unsupervised machine learning does not feature certain outputs, which signifies its purpose is to infer the occurring structure present in any type of data sets.

Supervised learning

Usually, supervised machine learning relies on classification, when it is the case of mapping the input to output labels, or on regression, when the sole purpose is to map the input to a continuous output. Hence, the most popular algorithms used in supervised learning include logistic regression, naïve Bayes, support vector algorithms or artificial neural networks. However, both regression and classification have a common purpose: to identify certain relationships or structures in the input data, which will permit us to effectively obtain correct output data.

When you are leading supervised learning, the primary considerations are represented by the complexity of the model, along with the bias-variance tradeoff. These two are closely linked, which signifies that supervised learning cannot be done without them. Model complexity refers to the specific function you are trying to learn, while the bias-variance tradeoff stands for the generalization of the model, obtained through the balance between bias and variance.

Unsupervised learning

Unsupervised learning represents a useful tool when it comes to exploratory analysis due to the fact that it has the ability to instantly identify the structure in the data. Some popular algorithms for this type of learning are k-means clustering, main component analysis or autoencoders. Thus, no labels are established, which means there is no clear manner to compare model performance in the majority of unsupervised learning methods.

Mainly, unsupervised learning algorithms are useful to pre-process the data while exploratory analysis is conducted or even to pre-train supervised learning algorithms, while supervised learning is mostly employed for exporting systems in image recognition, forecasting, financial analysis and so on.

Bottom line, opting for one of those two learning machine methods is closely linked to the factors that depict the structure and volume of your data. When it comes to real life, in the majority of cases both supervised and unsupervised machine learning are used together in order to get an accurate solution to solve the use case.