London faces deep social disparities, especially in housing. Low-income individuals are often excluded from both private rentals and homeownership. Gentrification intensifies this issue, as rising housing costs displace long-term residents in once-affordable areas.
This project explores whether patterns in housing or architectural design, as reflected in residential planning applications, can help predict neighborhood gentrification.
The first stage uses Natural Language Processing (NLP) with transformer models (BERTopic structure) to extract features from free-text planning application descriptions provided by the Greater London Authority (GLA).
The second stage predicts gentrification levels using GCN, LSTM, or Transformer (future suggestion) models, combining extracted features with external datasets (census data) across LSOAs.