Machine Learning and Cellular Automata: Applications in Modeling Dynamic Change in Urban Environments Open Access
Downloadable ContentDownload PDF
There have been several studies advocating the need for, and the feasibility of, using advanced techniques to support decision makers in urban planning and resource monitoring. One such advanced technique includes a framework that leverages remote sensing and geospatial information systems (GIS) in conjunction with cellular automata (CA) to monitor land use / land change phenomena like urban sprawling. Much research has been conducted using various learning techniques spanning all levels of complexity - from simple logistical regression to advance artificial intelligence methods (e.g., artificial neural networks). In a high percentage of the published research, simulations are performed leveraging only one or two techniques and applied to a case study of a single geographical region. Typically, the findings are favorable and demonstrate the studied methods are superior. This work found no research being conducted to compare the performance of several machine learning techniques across an array of geographical locations. Additionally, current literature was found lacking in investigating the impact various scene parameters (e.g., sprawl, urban growth) had on the simulation results. Therefore, this research set out to understand the sensitivities and correlations associated with the selection of machine learning methods used in CA based models. The results from this research indicate more simplistic algorithms, which are easier to comprehend and implement, have the potential to perform equally as well as compared to more complicated algorithms. Also, it is shown that the quantity of urbanization in the studied area directly impacts the simulation results.