Showing posts from February, 2012

Predictive maintenance

Machinery and equipment are typically serviced on a scheduled basis, determined either by the time lapsed since the last service or by the usage since the last service.  But is this the most effective way to service equipment?  If despite regular servicing, a machine fails, then costs associated with unscheduled failure are incurred.  And even if machines do not fail on a unscheduled basis, costs incurred on regular, scheduled maintenance may be more than necessary.

To explain this further, let us use an example.  Let us take the case of a simple machine that has three components A, B and C.  Let's say part A has a life of 3 months, part B has a life of 4 months and part C has a life of 5 months.  In order for the machine to keep functioning, we would need to replace part A every 3 months, part B every 4 months and so on.  This would mean that after the first three months, we would need to bring the machine down every month to replace at least one part!  In order to avoid this, on…

Claims denials: predicting acceptance / denial of claims

In today's post, we develop a predictive model that could be used to predict whether a medical expense claim will be accepted by an insurance company before the claim is made.  In order to do this, we will use the IBM SPSS Modeler data mining workbench to understand underlying relationships in observed data to train the predictive model.  The steps we will follow are as follows:

1) Understanding the data: among other things, this step will include determining which fields in the data to use as predictors and which ones to discard.

2) Partitioning the file: this step can be done within IBM SPSS Modeler by using the Partition node.  However, for today's post, we will do this in excel prior to creating the model.

3) Training the model: we will use one of the partitioned files to train the model.

4) Scoring the model: we will use the other partitioned file to score the model.

Let's get started.

Understanding the data

Since our data is in an excel file, we start with an excel so…