Modeling Impact of Transportation Infrastructure-Based Accessibility on the Development of Mixed Land Use Using Deep Neural Networks: Evidence from Jiang’an District, City of Wuhan, China

Mixed land use (MLU) plays a crucial role in fostering a sustainable urban development, vibrant communities, and efficient land utilization, providing a viable solution for smart growth, inclusive public transit, and urban sustainability. This study employs deep neural network (DNN) models: multilayer perceptron (MLP), and long short-term memory (LSTM), to analyze the effect of the transportation infrastructure-based accessibility on the prevalence of MLU patterns, based on the following data: infrastructure-based accessibility measures represented by the logsum (or transport supply), MLU patterns at the parcel level, and floor space prices by space type, for the years 2012 and 2015. Furthermore, the proposed methods are applied to the Jiang’an District of the city of Wuhan, China, at the parcel level as the case study. The study results reveal that MLU is predominantly accessible in areas close to the city center, characterized by a high density, and is relatively scarce on the city outskirts. Notably, parcels exhibiting mixed residential–commercial and residential land-use patterns underwent significant changes between 2012 and 2015, particularly in regions with robust accessibility via non-motorized modes and public transit, specifically in the central and southern parts of Jiang’an District. This transition is evident under scenario 3 (walk, bike, bus, subway) and scenario 6 (walk, bus, car) considered in this study. Furthermore, the study observed a substantial expansion in mixed commercial–residential and commercial districts, significantly near the high-transit accessibility area of subway line 1, as demonstrated in scenario 7 (bike, subway, taxi). The results from the MLP models show a mean relative error (MRE) of 4.7–14.08% for the MLU, and the LSTM models show an MRE of 3.74–10.38% for the MLU. More importantly, both the training and forecasting errors of the above models are lower, in most cases, than those reported in the literature. Moreover, these results indicate that the transportation supply or the infrastructure-based accessibility (represented by logsum) significantly influences MLU patterns.