Multi-Model Architecture Evaluation for Amenity-Based Price Prediction in the Kingston Region

Abstract

Municipal urban planners often face difficulties in determining the optimal location for developing city infrastructure due to the enormous number of contributing factors in evaluating economic ramifications. In collaboration with the City of Kingston, we determined property valuations to be strong indicators of the economic activity of an area. We introduce a dynamic multi-model architecture evaluation to assist our client, the City of Kingston, in developing a spatial econometric modeling system for urban planning purposes. Our approach evaluates four machine learning models trained on residential property transaction data using location-based attributes as predictor variables. Spatial clustering and logarithm functions were incorporated to capture non-linear relationships commonly observed in housing prices. Overall, our success metrics demonstrate that location-based factors play a significant role in determining residential property values, particularly in university-centered cities like Kingston. This project provides municipal staff with a practical tool to help narrow down and refine planning scenarios, helping improve efficiency and strengthen evidence-based policy development in urban planning.