Introduction to Predictive Analytics on SAP HANA

In this lecture we will look at Single, Double and Triple Exponential Smoothing. We will use Smoothing to make predictions using real-world data on Sea Surface temperatures. This lecture will cover:

• Loading of raw data into SAP HANA
• Transforming the data into a usable format
• Create the SQL for exponential smoothing procedure,
• Run the Seasonality Test procedure to get two of the required parameters for triple Exponential Smoothing
• View the results against actual observations for all three algorithms

All the SQL for the lecture is contained in the downloadable material below.

In the second part of Exponential Smoothing we will look at the HANA Analysis Process (AP) in SAP BW. Not only are we going to create the HANA Analysis Process in BW, but we will create a data flow to use the table we created in HANA. In more detail, we will:

• Set up HANA Smart Data Access
• Create a view in the Schema of the BW system
• Create an Open ODS view on top of the view
• Create a Composite Provider to act as a infoProvider to the HANA AP
• Create the HANA AP for Triple Exponential Smoothing and 
• Display the results in an Analytical Index

Why are our best and most experienced employees leaving prematurely? We will try to predict which valuable employees will leave next. 

In this lecture we will do some basic data exploration to get a feel for the dataset and also visualize the data in Lumira.

Generally, before we start the analysis of data, we have to do some data preparation.In this lecture we are going to do binning of the last performance rating into 5 categories. This will greatly reduce the complexity of the decision trees we are going to build in the next lecture.

This lecture discusses how decision tress are constructed. This lecture is optional and can be skipped if you are familiar with the math of decision tree construction.

Now that we have prepared our data, we will look at the various decision trees available to us. In particular we will look at:

• CART decision trees and
• C4.5 Decision Trees

Once we have completed the theory, we will run a small sql file to verify the first level split of both types of decision trees.

Note: Only the construction of the first level of the tree is covered in this lecture. The rest of the levels are left as an exercise and the answers are in the downloadable material for this lecture.

Now that we have prepared our data, we can use the full dataset to construct the decision trees. We will use the CART tree in this lecture to see which of our high performing employees are at risk of leaving. As usual the sql can be found in the downloadable material. Not only is the script for the CART tree attached, but the script for the C4.5 tree is also attached and you can run it to contrast the results of the two procedures.

This lecture will contrast the results of the decision tree when defining data as either continuous or discrete intervals. In the data preparation lecture we binned the last evaluation column into 5 discrete bins. In contrast, what would the effect be if we incorrectly classify Hours Worked as categorical?