User Modeling and Anomaly Detection [R]

Question

I have historical data of a online shopping website which is like;

userID        date        timeonSite   IsAnomaly?
   123   01/01/2012 12:38      14          1 
   548   01/01/2012 19:56      12          0

Its millions of rows and growing everyday too. Using R and MSSQL.

Trying create ml model for every users behaviour according to timeonSite value. And predict "Is next value an anomaly/outlier?"

Planned to logistic regression to use. But I need an incremental/online learning model for re-train users models every week or month. I can't use all of historical data every training session.

Is there any package to do this? Regression is not a red line, I am open for other methods like "ts, svm, knn etc."