What is Predictive Analytics?
Welcome back to Data Science Wednesday! On this week's episode, Decisive Data's Lead Data Scientist Tessa Jones teaches us how predictive analytics fits into the spectrum of analytics and data science.
What is Predictive Analytics?
Today we are going to talk about predictive analytics and what it can do for you and your business. Predictive analytics fits into the spectrum of analytics that we've talked about before, so first, let’s review a few key types of analytics before we move on to predictive analytics.
A Quick Review of Descriptive and Diagnostic Analytics
Starting with descriptive analytics, which is the most basic of the analytics, it's basically just cleaning, relating, summarizing, and visualizing your data. Descriptive analytics really gets to the questions about what is happening in your business.
Then there's diagnostic, which really gets down to why things are happening. What's causing my revenue to decline or to increase? How are things related? Things like that.
If you have a good base in both of those, then we are ready to move into predictive analytics, which is going to dive into what's going to happen in the future. If you're a business person and you want to be able to generate good business questions, if you have at least an idea of what might happen in the future, your answers are already going to be a little bit better.
Let’s start with an example because it just makes it easier to kind of flow through what's happening here. Pretend that we are grocery store owners. And if we're already talking about predictive analytics, you should have a pretty good grasp on descriptive, predictive, and diagnostic analytics—You probably already have a decent dashboard that really tells you what's happening in your business right now.
If you already have an aesthetically pleasing visualization that tells your revenue by different departments like foods and pastry, or how your sales changes by product over time, you have an idea of what's happening in your business.
One really common question is: how many of a given product am I going to sell for every store? Because this can really answer questions around how you're going to support supply chain processes, or how you're going to manage the profits that you're going to have.
What happened in the Past?
The first thing we need to do is talk about what happened in the past. We really can't do anything or predict easily unless we know or at least have an idea of what's happened in the past.
If we build a predictive model, it's going to tell us what's going to happen for the rest of the year, so if we build a model, we mix this information with all the past data that's really clean and well-organized, and we then mash it together with a bunch of mathematics and coding. Then we pop out some results and it shows up in a visual that shows the sales we have had and the sales we think we are going to have.
A business person can look at this visual and say, "Wow, we need to put a lot more products to this store because I see sales are going to increase." Or, "Our profit margins are going to be way higher than we thought so we can start a new program." You can really start to get innovative with your business decisions.
How well is it performing?
Let's pretend we've built this model and it's been running for a year. Now we want to know how well is this model actually performing? So down here, we have a chart that shows, in black, what we actually sold, and then in green, what we thought we were going to sell.
We see that there's a couple of pretty big misses, and so we need to go back and look at the data and understand what assumptions we applied that were maybe a little bit wrong, or applied incorrectly, or look at the data. Maybe we weren't accounting for something and we kind of reorganize that and incorporate it into the model. Then we would redeploy it, and have a better model.
This cycle can happen a couple times or it can happen many times. It really depends on the data and on the objective. It depends on a lot of different things. But we do try to minimize the number of times that we're having to iterate through this before we can have a really sound predictive model.
That is predictive analytics in a nutshell. Basically, once you have a solid foundation of descriptive and diagnostic analytics, we can really start pushing forward with predictive analytics.
Posted by Gage Peake