Evidence-Deficit Allocation for User-Centred Public-Space Measurement

Abstract

Public space analytics has become capable of capturing visual data, text data, behaviour data, environmental data, and administrative data. However, increased opportunities in evidence collection do not necessarily mean balanced evidence on users, since observable actions and simple perception tend to attract significantly more computational power than, for instance, safety, accessibility, climate comfort, universality, and management experience. The objective of this paper is to apply the proposed Evidence-Deficit Allocation approach to a ten-dimensional public space evidence register, featuring 427 dimension assignments and 58 dimension-level machine learning assignments. These ten dimensions include feeling towards place, satisfaction, sensory experience, use and activity, sense of safety, health, climate comfortability, perceived accessibility, universality, and feeling towards management. Evidence-Deficit Allocation involves several components including evidence share, machine learning share, local uptake, positive evidence-to-method deficit, and constraint load, which are then used to calculate a size-sensitive priority score and an under-adoption urgency score. Analysis finds that use and activity and feeling towards place represent 64.17% and 86.21%, respectively, of the total dimension evidence assignments, and machine learning assignments. It can thus be confirmed that there is a pronounced dominance of machine learning in behavioural and affective dimensions. Meanwhile, climate comfortability, universality, and feeling towards management represent 14.29% of dimension evidence assignment, but none of these three has any machine learning assignments. Use and activity achieves a maximum size-sensitive priority score of 100.00, while sense of safety scores 100.00 in urgency. Perceived accessibility, climate comfortability, management perception, and universality are four under-instrumented fields in public space evidence. In a 10,000-run perturbation analysis, the results are found robust against alternative constraint weighting.