Real life statistical analysis and data mining project case studies
Churn modeling for a nation -wide business newspaper
Data set size: 600,000 rows and 26 columns
Data base platform: MS-SQL 2008
A nation - wide business newspaper asked for assistance in its strategic plan to reduce churn rate. The newspaper’s analysts used Analysis Pro Logistic Regression suite on a well established data set. At the first phase of the project the software assisted the analysts to assign likelihood to churn score for each customer in the data set. In the second phase of the project the likelihood score formula was implemented as an SQL statement and was assigned to all of the customers. The churn score analysis has also reviled some of the major churn root causes. The reviling of the major churn causes led the marketing staff to change a part of the CRM flow.
Predicting User Value for a high traffic website
Data set size: 30 rows per affiliate
Data base platform: Excel 2007 based on Oracle 10g
A high traffic website uses the Analysis Pro Multiple Regression suite for predicting future user value per each affiliate. The web site has agreements with various affiliates that send traffic to it’s home and store pages. The price a website pays for every user sent by an affiliate is a major profit revenue factor. In order to optimize the amount of money paid to affiliates there is a need to predict the cash flow that a user produce for the website in the future time. In this case, the website asked to have a prediction for the next 365 days based on the generated revenue of each affiliate based on the first 30 days. The Best Regression Model” feature found per each affiliate revenues stream the appropriate regression type. The most common types were Power, Logarithmic and Exponential regressions. The total error interval was 2% - 15% for the first 90 days of prediction.