Predicting Local Innovation through AI and geospatial models: Revisiting the Innovation District Theory

This paper aims to deepen the understanding of variations in innovation performance across neighbourhoods within a city. Specifically, the study investigates the characteristics that influence innovation generation at the sub-city level to provide insights that would bridge the gap between national, regional, and local innovation policies. It aims to revisit the traditional Innovation District Theory in the post-AI era by proposing an innovation district model that shifts the focus from supply chains to innovation diffusion, moving beyond characterizing innovation districts solely by business metrics. To do it, we leverage an empirically merged dataset to explore well-established and newly introduced enabling factors of innovation generation. First, we employ a random forest algorithm to predict innovation performance using patent data from the census zip code tabulation areas (ZCTAs) across the New York metro area. This approach highlights significant neighbourhood assets associated with high innovation performance. Secondly, machine learning and deep learning models are trained and tested on ZCTAs from New York state with high predictive accuracy. Our findings show that when trained with neighbourhood assets and tested on New York state ZCTAs demonstrate high predictive accuracy, explaining 60% to 70% of the variation in innovation performance ten years later using specific neighbourhood asset variables. Additionally, when trained on data from the New York central metro area tested on data from Boston's central metro area, the accuracy remains relatively stable, suggesting an association between the assessed neighbourhood assets and innovation performance across these two US central metro areas. In the end, we introduce a newly constructed index, the Local Innovation Potential (LIP), designed to a neighbourhood’s innovation potential over the same period, in both states. These results reinforce the importance of key early neighbourhood-level characteristics strongly associated with high long-term innovation performance and provide a tool with practical implications for urban planning as well as for innovation policy design and implementation.