Prediction of Building Count and Dimensions from U.S. Census Data Using Multiple Regression

Abstract

Providers of high-speed, wireless data services need to know where in their service area to place transmitters to reach potential customers. Viewshed analysis, a technique found in Geographic Information Systems (GIS) software, can be used to model propagation of the wireless signals from different locations to find the best transmitter site. To carry out viewshed analysis, digital data are required for all obstructions the signal may encounter along its path. One such obstruction, terrain, can be represented in the GIS by easily available Digital Elevation Models (DEMs). Another obstruction is buildings, which are common in populated areas, and therefore of particular concern to wireless providers. Unfortunately, digital data for buildings in U.S. cities and towns are often non-existent, difficult to obtain, or very costly.

In light of the difficulties surrounding acquisition of building data for wireless propagation studies, this study used Multiple Regression analysis to construct models to predict building count and dimensions. U.S. Census Housing and Demographic data, aggregated at the Census Block level, served as the predictor variables in the regression equations. The models were built from sample data collected from four U.S. cities. For each variable to be predicted (Y), the top models were compared to find the optimum one. The model chosen for Building Count (per Block) showed quite good results, and future research in the prediction of this variable shows promise. Results for the models of Average Building Height and Average Building Footprint Area (both per Block) were not nearly as encouraging, but additional work modeling these variables may still yield insights.