IN THIS CHAPTER
Explaining the building blocks
Probing capabilities
Surveying the market
Predictive analytics is a bright light bulb powered by your data.
You can never have too much insight. The more you see, the better the decisions you make — and you never want to be in the dark. You want to see what lies ahead, preferably before others do. It's like playing the game “Let's Make a Deal” where you have to choose the door with the hidden prize. Which door do you choose? Door 1, Door 2, or Door 3? They all look the same, so it's just your best guess — your choice depends on you and your luck. But what if you had an edge — the ability to see through the keyhole? Predictive analytics can give you that edge.
Exploring Predictive Analytics
What would you do in a world where you know how likely you are to end up marrying your college roommate? Where you can predict what profession will best suit you? Where you can predict the best city and country for you to live in?
In short, imagine a world where you can maximize the potential of every moment of your life. Such a life would be productive, efficient, and powerful. You will (in effect) have superpowers — and a lot more spare time. Well, such a world may seem a little boring to people who like to take uncalculated risks, but not to a profit-generating organization. Organizations spend millions of dollars managing risk. And if there is something out there that helps them manage their risk, optimize their operations, and maximize their profits, you should definitely learn about it. That is the world of predictive analytics.
Mining data
Big data is the new reality. In fact, data is only getting bigger, faster, and richer. It's here to stay and you'd better capitalize on it.
Data is one of your organization's most valuable assets. It's full of hidden value, but you have to dig for it. Data mining is the discovery of hidden patterns of data through machine learning — and sophisticated algorithms are the mining tools. Predictive analytics is the process of refining that data resource, using business knowledge to extract hidden value from those newly discovered patterns.
Data mining + business knowledge = predictive analytics => value
Today's leading organizations are looking at their data, examining it, and processing it to search for ways to better understand their customer base, improve their operations, outperform their competitors, and better position themselves in the marketplace. They are looking into how they can use that information to increase their market share and sharpen their competitive edge. How can they drive better sales and more effectively target marketing campaigns? How can they better serve their customers and meet their needs? What can they do to improve the bottom line?
But these tools are useful in realms beyond business. As one major example, government law enforcement agencies are asking questions related to crime detection and prevention. Is this a person of interest? Will this criminal be a repeat offender? Where will the next crime happen?
Other industries, notably those with financial responsibility, could use a trustworthy glimpse into the future. Companies are trying to know whether the transaction they're currently processing is fraudulent, whether an insurance claim is legitimate, whether a credit card purchase is valid, whether a credit applicant is worthy of credit … the list goes on.
Governments, companies, and individuals are (variously) looking to spot trends in social movements, detect emerging healthcare issues and disease outbreaks, uncover new fashion trends, or find that perfect lifetime partner.
These — and plenty more — business and research questions are topics you can investigate further to find answers to by mining the available data and building predictive analytics models to guide future decisions.
Data + predictive analytics = light.
Highlighting the model
A model is a mathematical representation of an object or a process. We build models to simulate real-world phenomena as a further investigative step, in hopes of understanding more clearly what's really going on. For example, to model our customers' behavior, we seek to mimic how our customers have been navigating through our websites:
- Which products did they look at before they made a purchase?
- Which pages did they view before making that purchase?
- Did they look at the products' descriptions?
- Did they read users' reviews?
- How many reviews did they read?
- Did they read both positive and negative reviews?
- Did they purchase something else in addition to the product they came looking for?
We collect all that data from past occurrences. We look at those historical transactions between our company and our customers — and try to make consistent sense of them. We examine that data and see whether it holds answers to our questions. Collecting that data — with particular attention to the breadth and depth of the data, its quality level, and its predictive value — helps to form the boundaries that will define our model and its outputs.
This process isn't to be confused with just reporting on the data; it's also different from just visualizing that data. Although those steps are vital, they're just the beginning of exploring the data and gaining a usable understanding of it.
We go a lot deeper when we're talking about developing predictive analytics. In the first place, we need to take a threefold approach:
- Thoroughly understand the business problem we're trying to solve.
- Obtain and prepare the data we want our model to work with.
- Run statistical analysis, data-mining, and machine-learning algorithms on the data.
In the process, we have to look at various attributes — data points we think are relevant to our analysis. We'll run several algorithms, which are sets of mathematical instructions that get machines to do problem-solving. We keep running through possible combinations of data and investigate what-if scenarios. Eventually we build our model, find our answers, and prepare to deploy that model and reap its benefits.
What does a model look like? Well, in programming terms, a predictive analytics model can be as simple as a few if … then
statements that tell the machine, “If this condition exists, then perform this action.”
Here are some simple rule-based trading models:
- If it's past 10:00a.m. ET and the market is up, then buy 100 shares of XYZ stock.
- If my stock is up by 10 percent, then take profits.
- If my portfolio is down by 10 percent, then exit my positions.
Here's a simple rule-based recommender system (for more about recommender systems, see Chapter 2):
- If a person buys a book by this author, then recommend other books by the same author.
- If a person buys a book on this topic, then recommend other books on the same and related topic.
- If a person buys a book on this topic, then recommend books that other customers have purchased when they bought this book.