Abstract
—Local governments and lighting solution providers
are under pressure everyday to replace street lighting luminaries
with LEDs to save on the cost of operations and carbon
emissions. However, the correct specification of each replacement
fitting is a complex procurement task performed manually by
engineers and may easily result in excessive capital expenditure
or overspecification. This paper provides a framework that
implements machine learning (ML) using open datasets from
South Dublin County Council to automatically propose suitable
LED product specifications. Codification of past specification
policies was achieved through the training of tree-based classifiers
(Random Forest and XGBoost) on spatial coordinates as proxy
variables based on urban zoning. In addition, Shapley Additive
exPlanations (SHAP) are implemented to present procurement
rationales transparently at the asset level that are commensurable
with industry practices. Through this case study, the paper will
demonstrate how open public data can be utilised to provide an
interpretable, reproducible, and automated decision-support tool
to make products selection in large infrastructure development
programs effective.