Machine Learning is a scientific discipline in the field of Artificial Intelligence that creates systems able to learn automatically. Learning in this context means identifying complex patterns in millions of data. The machine that really learns is an algorithm that reviews the data and is able to predict future behavior. Automatically, also in this context, implies that these systems are improved autonomously over time, without human intervention. Let’s see how it works.
Big Data and Machine Learning applied to business
A telecom company wants to know which customers are in risk from unsubscribing their services to do business actions that prevent them from churning. How can it be done? The company has many customer data, many: seniority, contracted plans, daily consumption, monthly calls to customer service, latest changes in contracted plans … but surely, uses them only to bill and to make statistics. What else can be done with that data? It can be used to predict when a client is going to unsubscribe and manage the best action to avoid it. In short, with Machine Learning you are able to go from being reactive to being proactive. The historical data of all the clients, duly organized and treated in block, generates a database that can be exploited to predict future behaviors, favor those that improve business objectives and avoid those that are harmful.
That huge amount of data is impossible to analyze by a person to draw conclusions and even less to make predictions. Algorithms, on the other hand, can detect patterns of behavior counting on the variables that we provide and discover which ones have led, in this case, to unsubscribe as a client. The following image is an example of a simplified prediction based on data from a fictitious telephone company, but using a real Machine Learning tool:
The tree view (this image is simplified, the real prediction tree has many more nodes) allows to see the patterns that certain clients that have been unsubscribed have followed. In this case, one of the central branches is highlighted, which indicates a pattern in which the client:
- Made more than 3 calls to customer service.
- Calls less than 171.95 minutes daily.
- Calls made at night time last less than 189.02 minutes.
This is an analysis of historical data, but… where is the prediction? Let’s go to it: if customers who have these characteristics have already unsubscribed from the company, it is foreseeable that those who are still customers and have this same behavior are at risk of leaving. According to this predictive model, this is quite likely to happen (the prediction is said to have a confidence, in this case, of 91.97%). If the marketing department had this information, it could proactively propose a rate plan change or review why they have called customer service to try to keep them.
The complete prediction tree would be the following. In this case we have highlighted a false prediction (that is, it would not be written off) with a confidence of 90.59%. To the right of the image you can see the behavior pattern of this group of clients:
Why is it important?
The amount of data that is currently generated in companies is increasing exponentially. Extracting valuable information from them is a competitive advantage that cannot be underestimated. At CleverData we think it is an opportunity that should be given special attention. The great advantage is that currently it is not necessary to be a data guru to take advantage of this type of technology. There are tools in the market that are very easy to use (even for laymen in data analysis) and are economically affordable for any size of company that allows making predictions.
The challenge of taking advantage of the data has been greatly simplified. Today’s Machine Learning is not like before. This means that with quality data, appropriate technologies and favorable analysis, it is now possible to create behavioral models to analyze data of great volume and complexity. In addition, the systems provide fast and accurate results without human intervention, even on a large scale. The result: high value predictions to make better decisions and develop better business actions.
However, the volume of data should not distract our attention. It is not necessary to have as much data as Facebook or as a large bank to make models that help the business. It is better to have quality data (reliable and useful) than to have billions of data from which you cannot extract value. Start with something simple, use supervised Machine Learning, do not insist on using Big Data, use Machine Learning in the cloud and above all, start now. If you want, with us.
Machine Learning Application Fields
Many activities are currently taking advantage of Machine Learning. Sectors such as online shopping – have you ever wondered how the recommended products are instantly decided for each customer at the end of a purchase process? -, online advertising – where to put an advertisement so it has more visibility depending on the user who visits the web – or the anti-spam filters have been for time, taking advantage of these technologies.
The field of practical application depends on the imagination and the data that are available in the company. These are some more examples:
- Detect fraud in transactions.
- Predict failures in technological equipment.
- Predict which employees will be more profitable next year (the Human Resources sector is betting seriously on Machine Learning).
- Select potential clients based on behaviors in social networks, interactions on the web…
- Predict city traffic.
- Knowing the best time to post tweets, Facebook updates or send newsletters.
- Do medical pre-diagnoses based on patient’s symptoms.
- Change the behavior of a mobile app to adapt to the customs and needs of each user.
- Detect intrusions in a data communications networks.
- Decide what is the best time to call a customer.
Technology is here. The data too. Why should we wait to try something that can open the door to new ways of making decisions based on data? Surely, you’ve heard that data is the oil of the future. Now you can start pumping it.
Other articles that may interest you:
- The New York Times is looking to machine learning to help it understand reader behavior. via Gigaom.com
- Machine Learning is Fun!via medium.com
- Predicting CTR with Online Machine Learning. via mlwave.com
Original article (in Spanish): “¿Qué es Machine Learning?”
Translation: Sergio Paul Ramos Moreno
