It is a program to predict customers’ churn, appendence, or upselling behavior like KDD cup 2009. Raw data was cleaned with multiple steps. PCA was applied to reduce data dimensions. Multiple machine learning models were tried, and Gird search function was used to find the best parameters. ROC curves were used to summarize the results. See details in docx
I did several similar projects later than this one. See this link: https://github.com/skyblueutd/ChurnPrediction
There is a paper which summarized all aspects of churn prediction: http://research.sabanciuniv.edu/39116/1/AneelaTanveer_10236886.pdf