When we inside in data science and analytics world, we listening some strangers terms, like machine learning. But what really is machine learning? How a machine can learn? Where do I apply this?
Come with me, on this blog we will answer these questions!
Machine is present in almost all of your lives, you know that recommendations of friends in Facebook? So, Facebook use machine learning to determine who can be your recommendation to a new friend. Do you know the recommendation of movies that Netflix show us? So folks, this also use machine learning.
I listened a phrase at premiere of MIT Sloan magazine in Brazil, the speaker said “Artificial intelligence will be like tears in the rain”, that is, no one will notice but it is there.
With the machine learning utilization, the machines can do a lot of things, like play Mario Bros or chess, detect ill, classify objects and even drive.
Machine learning is an algorithm that receive the skill to learn with the data without being explicitly programmed or in a more technical way, it is a mathematical and statistical way of representing a business problem.
We have 3 types of learning:
- Supervised learning
- Unsupervised learning
- Reinforcement learning
In this learning we have the presence of a “teacher” saying to machine “Look, who has these characteristics is a cat, who does not have it is not a cat”
Following this logic line, in datasets we have variable X (characteristics) and the variable Y (label, target or output). In other words, from a new entry (X) I can predict my output (Y).
The problems involving this learning are divided into 2 categories:
*Classification: Given a dataset, with the target outputs (Y), the classification algorithm can determine the class to which a new input (X) belongs. For example classify whether the animal is cat or not.
- Regression: We use this algorithm when the target output (Y) is a real number, for example I want to predict the consumption of ice cream the next semester taking account temperature, day of the week and if the day it rained. It can be divided into simples linear regression: when I have an X variable, and multiple linear regression: when I have more than two X variables.
In this learning we don’t have the presence of a “teacher”, so the machine needs to learn alone the patterns and trends existing in the data, therefore, we don’t have the variable Y. The problems that involves this learning:
- Grouping: when we want to discovery groups in our data, for example to group people through the taste of books.
- Association: when we want to discovery rules that describe our data, for example how people who like book A, like to buy book B.
In this learning the machine learns through feedback on the results it has had and with this feedback the machine adjusts it behavior in order to seek the best result. This learning is widely used in games, drones and autonomous cars.
Here is some examples of applications with machine learning but there are many other problems that we can apply these algorithms too.
- Fraud detection
- Recommendation systems
- Search engines
- Demand prevention
- Predictive maintenance
- Diagnosis in healthcare
Until the next article !!
Writer: Laura Damaceno