TRAINING

This one-day class is designed for people who have large quantities of data and would like to become familiar with some of the advanced analytic techniques supported by InforSense. The examples and data used in class come from InforSense partner Data Miners, Inc., a consulting practice that specializes in analyzing data for marketing and customer relationship management purposes. However, the techniques themselves are equally applicable to scientific data.

A statistical model is simply a formal description of relationships that exist in data. Good formal descriptions have many uses. A good description of a profitable customer can be used to classify new customers as likely or unlikely to be profitable by measuring their distance from the prototypical profitable customer. A good description of who has responded to past offers can be used to predict who will respond to future offers.  A model may take the form of a set of rules or a mathematical formula. Either way, it can be tested for stability and accuracy so that it can be applied with confidence. This class will teach you what it takes to build stable models that remain effective for a long time and generalize well to new datasets.  Several popular data mining techniques will be introduced and demystified, including:

• Look-alike models
• Naïve Bayesian models
• Decision trees
• Linear and logistic regression
• Neural networks
• K-means clustering

These techniques will be applied to real data from real product penetration and churn modeling case studies. By studying the same business problems using several different modeling techniques, the class teaches a modeling methodology appropriate for all models while demonstrating the particular strengths of particular modeling approaches.

The class is primarily lecture, but students will have the opportunity to try out these techniques using InforSense.  The course material is based on the books Data Mining Techniques co-authored by the course instructor, Michael Berry, and Data Analysis Using SQL and Excel by Gordon Linoff.