SAP InfiniteInsight – Modeler
In my previous blog SAP InfiniteInsight – Explorer , I demonstrated how you can create a data set for further analysis.
In this blog I will focus on the SAP InfiniteInsight Modeler to create a model on the data set from my previous blog.
In the previous blog we prepared data that comes from a garden retailer that has a coffee shop. We prepared the data so we could analyze in this blog what will influence someone visiting the garden shop to most likely have a Dessert or Cake at the garden retailers coffee shop.
So lets start…
From the welcome screen I will select “Create a Classification” under the Modeler section. As you can see different types of models can be created.
I have now selected the data from the explorer. I have selected Analytical Record Set 1.
By pressing next you will go to the next screen, will be blank until you press Analyze button. Then figure 3 will be displayed. At this step we can also view the data if we need to.
Now we will select the target variable which we want to analyze, the target variable is who bought cakes or desserts. We also exclude some variables. So here we are saying we want the model to determine what the other variables impact is on our target variable.
The next screen will then show the summary of the model.
The model generation will start, also known as “Training the model”
The results of the model will be shown as seen in figure 7. It is important to know the following values shown and the meaning of the values.
- KI – a measure of how powerful the model is at predicting. This is a number that ranges between 0 and 1. The closer the KI is to 1, the more accurate the model is.
- KR – a measure of robustness, or how well the model generalizes on an independent hold out sample. KR should be a number ideally above 0.95.
So based on the above, our KI measure is poor. But will serve our purposes for the blog 😉
We can now review the model results by selecting the appropriate options.
By selecting “Contribution by variable” we can see that the following aspects influence the scenario. Firstly pets, then children, then the segment, then the age, etc.
We can now take it further and analyze the age variables. Here we see that ages between 18-26 and 48-70 are likely to buy a cake or dessert. Individuals with the age 26-48 are less likely.
So this now tells us the coffee shop will have better success with cakes and desserts that are appealing to people with pets, have children and are between the age gaps identified. This will help deliver a more precise advertising if needed.
Hope the above shows how a predictive model is created by just clicking away and how the results can be a valuable tool.