Understanding Residential Address Patterns in Urban and Rural Areas: A Machine Learning Approach

Address data quality has a direct impact on demographic and other spatial analyses, since it may lead to uncertainty and potential bias. Most of the existing studies measure address quality through matching with reference databases, which can be an expensive and time-consuming process. To bridge this gap, we propose a multiclass classification algorithm to evaluate the syntactic quality of residential addresses from a large database without using external databases. Namely, we adopt a multi-objective optimization approach, based on the NSGA-II algorithm and two modified k-NN algorithms. The objective is to find the address components as well as the optimal number of neighboring examples that help explain which class (good, incorrect or incomplete and anomalous) the quality of an address belongs to, by type of region (urban, medium urban, and rural). The presented results indicate that the proposed approach outperforms the best baseline algorithms on multiclass classification, while also providing descriptive information on the most relevant features and median local neighborhood of each instance. With this study, we further extend previous research in the field of address pattern extraction, by explicitly differentiating urban and rural areas as well as invalid and anomalous addresses.