Now, during the Third Industrial Revolution, the radical change is that we have algorithms capable of learning by themselves, either by training or cold data.
During the First Industrial Revolution, man’s manual capacities and strength were artificially replaced with the extraction of fossil minerals. During the Second Industrial Revolution production lines were developed, giving way to today’s cars with 250 horsepower and electric water pumps that extract water in farms for crops. Now we bear witness and participate in the Third Industrial Revolution.
During the Third Industrial Revolution, cars become intelligent to drive themselves, and water pumps are smart enough to extract the necessary amount of water considering reserve volume, meteorological conditions, the season and the crop.
In 1956, Arthur Samuel wanted his computer to beat him in a game of checkers. After many hours of trying to teach it the best plays, he decided to make it play against itself so that it learned on its own. In 1962, his computer won the Connecticut championship in the US. It is Samuels who we attribute with the creation of machine learning.
Artificial intelligence began from the knowledge of experts, meaning that a computer had to be told what to do in explicit detail. Now, during the Third Industrial Revolution, the radical change is that we have algorithms capable of learning by themselves, either by training or cold data.
Algorithms that classify or predict without previous knowledge are called “unsupervised”, and those that require training are called “supervised”. In supervised algorithms, training data consists of pairs of objects (usually vectors). One component of this pair is input data, and the other one, the desired results. The output of the function can be a numeric value (as in regression problems) or class tag (fraud/normal transaction/suspicious transaction).
An important characteristic that must be noted of supervised algorithms is overfitting. This happens when the algorithm is over-trained in the training set and loses the ability to generalize; this means that it becomes highly specialized in what is known in the training set, and its performance is very bad in any situation unforeseen by the training set.
An overfitted model has a low predictive performance as it overreacts to small fluctuations in training data. The models must generalize in unforeseen situations based on the presented data.
The automatic learning algorithms we have developed and incorporated in Monitor Plus® are:
Naive Bayes Clasiffier;
Online Data Mining;
Deep Learning Neural Networks;
These technologies are ingrained in the DNA of Plus Technologies and are present in the conception of Monitor Plus®, which is why throughout the company’s lifespan we have incorporated as many supervised as unsupervised learning techniques.
In conclusion, there is a premise that if a classifying algorithm is good, using several may be even better. In this sense, the techniques are assembled to generate a unique final classifier that delivers a better performance in detection levels and in false positive relations.