Supervised Machine Learning Approaches to Modeling Residential Infill Development in the City of Los Angeles

Abstract

While infill development is widely accepted by cities as an alternative to urban sprawl, a very dearth of research has attempted to measure infill development and identify contributing factors to infill development. Filling this research gap, this paper models residential infill development in the City of Los Angeles by employing five machine learning (ML) algorithms. This paper attempts to identify the best-performing ML algorithms by comparing the performance of the ML algorithms. Of the five ML algorithms tested, the random forest (RF) and k-nearest neighbor (kNN) algorithms are selected as the best-performing algorithms. The RF algorithm ranks independent variables from most to least important. Overall, the ranks suggested that residential infill development in the City of Los Angeles is significantly influenced by the physical conditions of property and neighborhood rather than socioeconomic characteristics. Diverse land uses, good housing mixes, and rail transit accessibility also, importantly, contributed to the infill development. This finding suggests that the city’s planning efforts, such as the promotion of accessory dwelling unit (ADU) development and the expansion of rail transit, can create a virtuous circle for sustainable urban development.