Machine Learning for Local Government: Using Maximum Entropy to Predict Parcel Development
Machine Learning for Local Government: Using Maximum Entropy to Predict Parcel Development
Abstract
How can machine learning help predict where development will happen next? This presentation applies Maximum Entropy (MaxEnt) modeling to evaluate development suitability across vacant parcels in Champaign County, Illinois. Using a binary classification of parcel codes to distinguish developed from undeveloped land, I generated presence and background points, then modeled suitability using four environmental predictors — population density, proximity to roads, distance to existing development, and topographic slope — drawn from local, USGS, and Census sources. Implemented in ArcGIS Pro’s Presence-Only Prediction tool (though open-source alternatives will be discussed) with a hinge basis function, the model achieved strong discriminative accuracy, with distance to existing development emerging as the dominant predictor. I’ll discuss what these findings suggest for local land-use planning, along with key limitations tied to parcel data quality and classification ambiguity, and close with directions for extending this suitability-modeling approach.