There’s no doubt that being able to predict how your clients will act, can give you an edge over the competition. WIth predictive APIs, the technology is now in place for organizations to realize such an advantage.
“We’re now talking about predictive apps and predictive-first development in the same way as we were talking about mobile-first development a few years ago,” said Louis Dorard, founder of the PAPIs predictive API event, at an APIcon last year.
This article attempts to help you understand how you can turn your data into insights with predictive apps and predictive APIs.
What is Machine Learning?
Machine learning (ML) is a kind of artificial intelligence or guided learning that allows you to get informative predictions from your data without having to go through the process of programming it all. The program has a built-in pattern recognition mechanism that learns behaviour by examples. Predictive apps are ML combined with data.
There are three main types of machine learning:
- classification: multiple choice answers, each corresponding to a class like email’s “spam vs. ham,” it uses predictive analysis to answer the question: “What type of email is this?”
- regression: numerical answers, like the value of a house assessed by a real estate agent. Take characteristics of houses on the market like number of bedrooms and bathrooms, overall size, and year built along with their pricing information. With machine learning, we are trying to fill in the missing values of new houses to market, based on the input you can give. The ML should give the pricing output.
- recommendation: Simply, given this user, what are the top five products to show him?
According to Dorard, machine learning takes example data when you have both inputs and outputs; it goes through a Training Phase in order to build a model. The machine learning doctorate says there’s another phase which is predicting based on a model and any new input—for instance, a new house or a new email—which offers a prediction on the output. You’re going to say whether the email is spam or not. You’re going to say how much the new house is worth.
Dorard says that “In this way, it’s quite similar to what we do as humans. If we want to give estimates of house prices, we would look at what’s on the market, we would look at several examples in our heads, and we would build a model of the dynamics between the relationships between the characteristics of real estate properties and their prices.”
Quantum physicist and founder of the MOCA Platform for mobile marketing Dr. Maria Fernanda Gonzalez would beg to differ, worrying that people will confuse predictive software with the more complete ML. She would argue that much of what is currently dubbed ML is really predictions. “Learning is much more complicated. Machine learning isn’t this easy because machine learning is about bringing together the different variables to predict what will happen.”
“Predictive is more sensible than machine learning. Predictive in traditional business intelligence is one-dimensional and machine learning is multidimensional,” Gonzalez continued, saying that ML is about bringing factors together, in a much more sophisticated way, incorporating much more difficult mathematical tools. From a maths standpoint, “Prediction is very easy—it’s just a functional fit.”
Applying it to her platform’s area, “Marketing is rudimental. You look at data and you interpret and use that data. The idea with machine learning is that the machine does it alone, experimenting,” with its given level of autonomy until it finds the predictive patterns that can combine to take your algorithm further so you can begin to learn from the data selections. For marketing, ML would be applied to not only find a pattern of customer behavior but to predict interrelated behavior patterns that come together as learned conclusions which can then be turned into personalized mobile marketing techniques that trigger the shopper in the buying cycle.
Apple’s Siri and Google Now are certainly at the brink of this machine learning, but one of the big names in ML Google’s DeepMind goes a lot further, as it tries to make machines learn for themselves. This will take AI to the next level of AGI or artificial general intelligence, where, instead of learning a specific task, machines start to gain human cognitive abilities like problem solving and reasoning. This means that machines could even become more adaptive and flexible for learning than us humans.