Within the field of Artificial Intelligence (AI) there is a branch called Machine Learning (ML), which allows predictions to be made based on data analysis. This technology has gained popularity in recent years, offering various solutions in different fields of knowledge, with a focus in the automation of processes, that reduce time and costs, through a machine that is capable of learn.
Next, we will explore the definition of Machine Learning (ML) and examples of some real applications, where you can see opportunities for your projects.
Arthur Samuel (1959) defines it as: ¨Field of study that gives computers the ability to learn without being explicitly programmed¨, in simple terms that a computer can learn by itself. An ML application that we are in touch, is the process in charge of classifying the email as SPAM or NO SPAM, in this task the algorithm reviews the emails and makes a prediction, which can be modified by the user, indicating that the email is not spam and in the opposite sense that it is spam. These user actions feed the algorithm to improve its next predictions and thus enter a cycle of continuous improvement in the task of classifying the mail.
There are 2 main types of algorithms in ML:
- Supervised learning, consists of supplying groups of data to train a model, that include the desired results to allows the model to measure its accuracy, and execute refinement cycles until the error is minimized. The 2 types of problems that this algorithm can manage are classification problems, such as identifying if there is a cup of coffee in an image, and regressions, which are commonly used in predictions, such as the housing price, based on characteristics. of the property.
- Unsupervised learning, analyzes data without any particular reference or desired result as in supervised learning, so that the objective is to discover relationships or patterns between the data, an example of the use of this algorithm is in the Google News application, which classifies and related news articles from different sources.
Case studies in ML implementation
This company is dedicated to the generation and distribution of electricity, for which it has several power generation technologies, among which is a wind farm with many generators, and each one must be visually inspected to determine maintenance tasks.
The task of inspecting is done on images taken by a drone of the generator blades, which for the entire field can be around 30,000 images. This task requires a large amount of time, since each image must be evaluated by a person.
The AES proposal was to use ML to train a model that reviews the images and identifies damage in the generators, so that the algorithm classifies the images, reducing the number of images that a person must review by half, since it removes images that shows no damage in the generators.
This credit and debit card franchise tracks transactions for fraud using ML, in a combination of supervised and unsupervised learning, the former algorithms learn about fraud that has occurred and the latter to detect anomalies in the data, and determine if a transaction was a fraud.
This movie and series platform offers more than 14,000 options and for its users to quickly find something they like, it has a recommendation system that uses ML. This algorithm will evaluate the categories visited, movies and series watched, as well as profiles of users with similar tastes, to make recommendations for movies or series, thus providing a personalized and satisfactory experience.
ML has had a great development in recent years and added to the large amounts of data that we collect and have available; models can be trained that offer companies of all kinds possibilities to partially or totally automate their processes. These technologies can help you solve problems in a new and innovative way, reducing time and costs.
If you want to know more about this technology, at DEJ Software we can advise you, we invite you to contact us.
AuthorJuan Carlos Valderrama GonzálezWeb/mobile development team Leader - DEJ Software