There is ample evidence of the role of land use and transportation interactions in determining urban spatial structure. The increased digitization of human activity produces a wealth of new data that can support longitudinal studies of changes in land-value distributions and integrated urban microsimulation models. To produce a comprehensive dataset, information from various sources needs to be merged at the land-parcel level to enhance datasets with additional attributes, while maintaining the ease of data storage and retrieval and respecting spatial and temporal relationships. This paper proposes a prototype of a workflow to augment a historical dataset of real estate transactions with data from multiple urban sources and to use machine learning to classify land use of each record based on housing market dynamics. The study finds that engineered parcel-level attributes, capturing housing market dynamics, have stronger predictive power than aggregated socio-economic variables, for classifying property land use.