Machine-Learning Reproducibility Package for the Cultural Stations Survey

Jelena Atanacković Jeličić, Dejan Ecet

This reproducibility package contains the datasets and Jupyter notebooks used for the machine-learning analyses reported in the study <i>The Unexpected Consequences of Decentralization: How Cultural Infrastructure Reorganizes Urban Attractiveness</i>. The research examines accessibility, participation, urban attractiveness, symbolic significance, and behavioural patterns within the decentralized Cultural Stations network in Novi Sad, Serbia.The package includes three independent analytical workflows: multi-class classification of perceived urban significance, binary classification of local versus non-local behavioural choice, and unsupervised clustering of user typologies. It contains three CSV datasets, three corresponding Jupyter notebooks, and a README file describing the variables, file dependencies, software environment, and reproducibility procedures.The machine-learning analyses were conducted in Python 3.11.7 using PyCaret 3.3.2, with a fixed random seed (session_id = 1971). The files reproduce the final machine-learning analyses used in the study and are provided to support transparency, verification, and reproducibility.