Machine learning is a branch of computer science that includes methods and algorithms. Those aim to make a machine capable to learn without being explicitly programmed to respond to a certain questions or to perform certain operations.

Machine Learning is closely linked to the statistical calculation. It’s often used in the data analysis to make predictions, as well as in the artificial intelligence in which it was born.

With machine learning it’s possible to deal with three main issues categories:

  • Supervised learning: a number of input and output examples are provided to the machine. This machine, has to identify and learn the rules that link them to each other, to be able to provide outputs to new inputs. In this category there are classification and regression iusses, whose goal is to entrust inputs to already known classes. Inputs are moderate in classification and continuous in regression.
  • Unsupervised learning: only a set of inputs are provided to the machine from which to extract pattern and structures that link them. In this category there are clustering issues, whose goal is to group a set of inputs into classes not previously known. The algorithm will have to discover by itself the recurring patterns.
  • Reinforcement learning: the machine interacts with an external and dynamic environment in which it has to accomplish a given goal. For each action that it does, the system gives a positive or negative feedback in relation to the goal to be achieved. From these feedbacks, the machine learns. Reinforcement learning finds application in many IT and statistics branches, such as game theory, genetic algorithms or in different industry as economy.

Therefore machine Learning is a set of powerful and still evolving tools, practically applicable to every field of human activity, usable to better understand the data at our disposal, extracting new information.

Together with the machine learning we want to introduce the predictive analysis concept too, as a set of techniques mainly deriving from it and also from data mining. Starting from an existing set of data, machine learning and data mining allow to extrapolate patterns to perform projections on future behaviors, evaluating the fields they come from. Thanks to the predictive analysis it’s possible to find frauds and failures, to evaluate risks and to choose marketing strategies. There are multiple applicative environments, like economy, tourism, telecommunication and health care.

Various machine learning techniques are used with predictive analysis, like classification and regression algorithms, clustering, neural networks which, through appropriate metrics, allow to evaluate the answers of the required predictions too.

Even if all this field is about really powerful tools, it’s crucial to understand that the truthfulness and the goodness of the achieved results is still closely linked to the validity of the data given as input. For that reason it’s strategical to start from a phase of data analysis and clearance before executing the chosen algorithm.
Therefore we can assert that data analyst is strictly necessary as well as an in-depth knowledge of the data and the environment in which we are operating.