Machine Learning explained for everyone
In this new post we are going to explain in a simple way what Machine Learning is, what is this term that is very fashionable and more and more present nowadays.
When it comes to explaining a concept, telling how a certain technology works or selling a service, it is extremely important to leave technical language aside and convey our ideas in a simple way. We could say that it is a friendly, and why not inclusive, way for people with little or no knowledge to understand what we do. In turn, this facilitates word of mouth, so having the ability to simplify complex concepts is key. That is exactly what we will try to do in this blog when talking about Machine Learning.
What is meant by using Machine Learning?
Why do we talk about " learning"?
In a simple way, the computer is able to "learn" from data provided by the user in order to achieve a specific goal. The way the computer achieves this learning is through algorithms. Algorithms are a series of defined steps that are programmed in a certain language that the computer will be able to interpret and execute. An algorithm can range from a simple equation of a straight line, to complex systems involving statistical methods, mathematical rules and logical expressions. The purpose of implementing these algorithms is to give the computer the ability to interpret the data provided to it. Behind this is the intention of obtaining valuable results, being able to find patterns hidden to the human eye, and even process a larger amount of data than one would be capable of.
What types of Machine Learning are there?
The problem to be solved will define the environment in which the solution will be developed and the factors that will affect decision making. Because of this, we can find different types of machine learning: supervised, unsupervised and reinforcement learning.
Supervised machine learning
Here the data with which we build the model are "labeled". This means that the outcome of the variable we want to predict (dependent or target variable) is known for certain observations. For example, if we want to train a model that predicts whether a customer will cancel his subscription, we will need a dataset with a variable containing the outcome of the cancellation (cancelled or not cancelled) for previous or existing customers. This outcome has to be labeled by someone before training a model. If this dataset contains 5,000 observations, it is necessary that all of them have the labeled outcome. The goal of the model is to learn the relationship between this outcome column and the other characteristics (also called independent variables or predictor variables).
Unsupervised machine learning
It is the opposite of supervised machine learning. Our data are not labeled and these labels are not necessary to build the model. Within this type of algorithm, clustering is the most commonly used. Clustering is very popular for customer segmentation. Here the algorithm seeks to group customers with similar behaviors and/or characteristics from the data provided.
Los algoritmos de aprendizaje por refuerzo se distinguen aún más que los dos anteriores. Aquí se definen modelos y funciones enfocadas en maximizar una medida de “recompensas”, basados en “acciones” y el ambiente en donde se desempeñará el sistema. Este algoritmo es el más relacionado a la psicología conductista de los humanos. Es un modelo acción-recompensa, en dónde el algoritmo busca la mejor “recompensa” dada por el ambiente, estando sus acciones sujetas a estas recompensas. Este tipo de métodos pueden usarse para hacer que los robots aprendan a realizar diferentes tareas cómo por ejemplo: jugar al ajedrez. Aquí el robot aprende jugada tras jugada a partir de la “recompensa” obtenida en cada una de ellas. Esta recompensa se define de acuerdo al objetivo del sistema. En el caso del robot, la recompensa podrá ser la probabilidad de ganar la partida.
It only remains to say that the approach chosen will depend on what you want to achieve. Each type of machine learning has its advantages and areas of opportunity. The challenge lies in knowing how to determine the nature of our problem. In this way, we will be able to determine which type of Machine Learning is the most convenient to implement and, in this way, succeed in solving our problem.