One of the common analytics models you learn about in Chapter 7 and build in Chapter 11 is regression. Don’t worry about the name right now (or the math). Regression is kind of like calculating the slope of a line on steroids. A regression model basically examines your data and figures out a line (or a curve) that matches the data you’ve seen. After you can graph your data, you can use that graph to guess what will happen based on new input data.
Let’s see how that can help. Figure 1-2 shows a linear regression model built on audition data and resulting score data. This example comes from an example you use to build this model in Chapter 11.
FIGURE 1-2: Linear regression model using hours practiced and audition scores data.
Here’s the explanation you see again in Chapter 11: Suppose you're helping student musicians prepare for honor band tryouts. You've collected historical data on how many hours a week each student practiced, whether the student was accepted in the honor band, and what audition score each student earned. As you would expect, a linear correlation exists between hours of practice and audition score: The more a student practiced each week, the better score that student earned at his or her audition. A linear regression model can predict any student’s audition score if you know how many hours that student practices each week. If you have a student who practices 30 hours per week, you could expect that student to earn a score of about 60 on the audition.
Regression models can help to accurately predict future actions. Using data to know what’s next can be worth its weight in gold when making business decisions. (Yeah, I know data doesn’t have weight, but you get the point.)
Making decisions based on models
Analytics models can help organizations make astounding decisions and gain lots of money. They can also lead organizations to make dumb decisions and lose lots of money. The trick is in knowing how good your models are.
This book is about building analytics models using blockchain data. You learn about blockchain technology and data in Chapters 2 and 3, but don’t forget that although the quality of your data is important, building the right model is crucial to getting quality output. Never rely on your first choice of a model or on a single model. Always compare model types and configurations to find the right combination to return the highest quality results.
Changing Business Practices to Create Desired Outcomes
Classifying your customers or building models to predict what comes next can help your organization be more responsive to needs. You can use analytics to help plan better and be ready for whatever comes next. But with some additional work, you can do far more with analytics results. Instead of just getting ready for what might happen next, you can use analytics results to alter today’s activities and affect future outcome.
Predictive analytics predicts what future results may be. The next step in analytics maturity is prescriptive analytics. With prescriptive analytics, the model identifies changes you can make now to achieve a desired outcome. For example, prescriptive analytics can tell you how many tables to set out in a restaurant or which register lanes to open in a grocery store to meet sales goals. Prescriptive analytics gives organizations the leverage to make operational changes based on their understanding of data that leads to satisfying their goals.
Defining the desired outcome
In the preceding section, you learned about using analytics models to make predictions of future outcomes. There can be tremendous value in prediction, but you can use analytics also to set the outcome and tell you how to get there. Think about it. It's one thing to predict next week’s sales, but wouldn’t it be cool to set your next week’s sales goals and let your analytics models tell you how to get there? With good analytics models, it's possible.
Predictive analytics basically gives you an equation: y = mx + b (yes, that’s a simple one and the same as the point-slope form of a line). Your model provides values for m and b. Your data provides a value for x and you solve for y. Simple algebra.
Prescriptive analytics is a little different. Prescriptive analytics ask the question: “If I choose a value of y, what value of x will get me there?” In other words, you choose a value of y (maybe your goal for next week’s sales), and then solve for x. After you know x (perhaps x represents the number of prospect calls you need to make), you know what it will take to reach y (your sales goal). At its core, it's still simple algebra.
Even though the algebra is simple, putting prescriptive analytics into practice can be tricky. In algebra, equality is reflexive, which means you can read left-to-right or right-to-left. Technically, models should work the same way, but they don’t always work that simply. Prescriptive analytics can provide some guidance on reaching goals, but you always have to take that guidance with a grain of salt. Try your model’s recommendations, and then evaluate the results. Fine-tune your changes, and then try it again. The best use of prescriptive analytics is as a good suggestion, not a surefire approach to reaching goals.
Building models for simulation
One of the challenges in prescriptive analytics is the iterative and flexible nature of using models this way. Predictive analytics is pretty straightforward. You can determine future outcomes within a known range of error. When turning that model around and using it for prescriptive purposes, you can never be sure that your model is taking into account all the influences that affect outcome. The outcomes your predictive model measures may include unsampled features (characteristics) that happen even though you don’t measure them. If this is the case, just changing one feature may not have the effect you expect.
Because prescriptive analytics is more than just turning a predictive model backwards, you’ll have to run your model multiple times over your dataset, changing a single feature at a time. Building a model that is flexible enough to respond to multiple feature changes is the basis of simulation. You're simulating the nature of reality, which encompasses multiple features that change and some level of unmeasured uncertainty.
Investing the effort to build a good simulation can more than pay for itself. A solid simulation is flexible enough to change as new input shows different trends and still provide output that you can trust. A simulation that tells you how to reach your goals is even better than knowing the future.
Aligning operations and assessing results
The best response to having good analytics models is to change operations based on your model output. Whether your focus is understanding your business, predicting tomorrow’s environment,