Application of Classification and Regression Trees for Paging Traffic Prediction in LAC Planning

Automatic methods for location area code (LAC) planning in mobile networks require prediction values of the expected signaling traffic (paging, location update, etc.) that have to be provided by traffic and mobility models. Modeling can be based on a learning data set consisting of geographic information (population distribution, land use) as input data and performance measurement values from the operations and maintenance center (OMC) as output data. In this paper, the performance of fast modeling methods based on classification and regression trees (CART) is investigated and compared to linear regression analysis. It will be shown that a combination of these two methods shows modeling results of arbitrary accuracy. The analysis of the modeling performance is carried out by comparing the mobile terminated call (MTC) prediction values with OMC measurement values from a real network.