A data-driven approach for land-use configuration

Abstract

Efficient land-use is critical to support human activities by building up the corresponding eco-system. Land-use planning for the rapidly changing world is a perpetual problem [1]. With the increase of population, land-use configuration plays a more and more important role to utilize land resources more efficiently and obtain a more beneficial outcome for the society and environment. Traditionally, urban planning is decided by experts based on their specific knowledge [2]. However, Judgment solely based on experts’ subjective criteria through traditional methods like an offline manual survey is time-consuming and costly to make the best choice for land-use configuration [3]. In this final year project, we use a data-driven approach to promote land-use configuration, reducing cost and improving efficiency. Since Google Maps is a widely used mapping service, which can provide us a great deal of useful information about sites including location, surrounding details, and visitors’ comments, we take the data collected from Google Maps as criteria to judge whether the decision of land-use is desirable. Since few studies have been done to address this task, this report considers four fundamental machine learning models as the baselines and try to find the best one among them. Specifically, we conduct four different fundamental models, namely the linear regression, decision tree, logistic regression, and deep neural networks. We apply the suitable data pre-processing strategy and then train these models. Finally, we discover the decision tree method has the best result and thus we treat it as a solid benchmark for facilitating the future research in land-use configuration.