Models

We need to create a predictive model that can be used to simulate an outcome.  This simulation will later be used to consider the various strategies that will be needed to achieve our objective. A predictive model is created using regression analysis and other techniques that make use of business intelligence technology.

Regression analysis uses a mathematical algorithm to make predictions based on the drivers that affect our outcome.  When we create a predictive model we use the same dataset that was used for correlation analysis. This is referred to as a training dataset. The model that we create should be able to accurately predict the outcomes that are found in the training dataset.  An algorithm can be adjusted to increase accuracy.  In addition, there are multiple algorithms that we can choose from.

The example that we use to explain the Weincor business intelligence method is as follows (1)

Using regression analysis we create a model and determine it's accuracy (column Percentage Correct) to predict a student's decision to accept or reject an offer of admission

Our predictive model show an overall accuracy of 94.4%.  The ability to predict a student's decision to accept an offer of admission is 98.2% while the ability to predict a decision to reject an offer is 69.2%.  The number of rejected offers in our dataset is small and a larger volume may have an affect on accuracy.  Further configuration of the algorithm or possibly another algorithm may also improve accuracy.

In addition to a predictive model, we create a simulation that allows us to reproduce the contents of our training dataset based on the original distribution of values.  If our predictive model is accurate the simulated dataset will yield the same student decisions that were found in our training dataset.

We run a baseline simulation and the following rate of accepted and rejected offers of admission is observed.

The distribution of values for each driver will be adjusted based on the outcome that we want to achieve in our target simulation.  Click on Strategies to learn how we determine which drivers to adjust, the distribution of values that will be needed and identify the strategies that will yield the target distribution.

(1) The data that was used to perform the regression analysis and baseline simulation was included in the book Dr. Green's New Hyper-Linked Introduction to IBM SPSS Modeler, Dr. John B. Green Jr. PHD, Left Brain Books, LLC. The successful use of business intelligence depends on the proper identification and collection of data.  Dr. Green provides a valuable data resource to illustrate our method.