Monday, August 23, 2010

Predict 3G users for telecom by using SAS Enterprise Miner


Situation: For a telecommunication company, there are a training dataset of 18,000 customers and a scoring dataset of 2,000 customers.
Task:Find potential 3G users from the existent 2G users to increase ARPU and MARPU
Action: Trained models by decision tree, neural network and logistic regression on SAS EM 5.2.
Result: Proposed tailed device and service, promotion channel, and branding image strategy for segments; Formed an ensemble model with misclassification rate <.04 and Impremented the model.

/*VERY BEGINNING: DATA TRANSFER FOR MODELING BY SAS ENTERPRISE MINER*/
options noxwait noxsync;
dm 'x "cd D:\";';
dm 'x " md mylib" ';
dm 'x "xcopy d:\matchresult\*.*/D/E/S/R/Y/A d:\mylib " ';

Good math, bad engineering

As a formal statistician and a current engineer, I feel that a successful engineering project may require both the mathematician’s abilit...