Machine Learning for Predicting Community Displacement Risk: Policy Transition from Reactive Intervention to Proactive Community Stability

This article introduces an open-source project that uses machine learning to predict community displacement risk. Its core goal is to help urban planners shift from reactive response to displacement issues to proactive prevention, enabling long-term stable development of communities. Taking Charlotte, USA as a case study, the project builds predictive models through data-driven methods to support precise policy-making, promoting social equity and urban sustainable development.

Charlotte is one of the fastest-growing cities in the southeastern United States, but rapid growth has exacerbated housing affordability challenges: housing prices have grown far faster than income, gentrification pressure has increased, minority communities face higher displacement risks, and infrastructure investment is uneven. In response to these issues, the Charlotte municipal government and community organizations urgently needed tools to identify risks, leading to the birth of this project.

Data Collection and Feature Engineering

Integrate multi-source data: demographic statistics (income, ethnicity, family structure), housing market (housing price trends, rent, vacancy rate), infrastructure (transportation, schools, medical care), historical investment, and community stability indicators (length of residence, organizational activity).

Model Selection and Training

Explore multiple models: logistic regression (baseline, high interpretability), random forest (captures non-linear interactions), gradient boosting trees (XGBoost/LightGBM, excellent performance on tabular data), and neural networks (potential for spatiotemporal sequences). Time-series cross-validation is used to ensure generalization, and model fairness is emphasized to avoid systemic bias.

Risk Scoring and Classification

Output community risk scores, divided into four levels: low, medium, high, and extremely high risk, to help policymakers allocate resources according to priority.

Early Warning Signals

• Housing prices have grown above the city average for consecutive years
• Demographic changes (increase in the proportion of young professionals, changes in ethnic composition)
• Shift in business ecosystem (small shops replaced by chain brands)
• Infrastructure improvements (new transportation lines, park renovations, etc.)

Spatial and Temporal Patterns

High-risk communities show spatial clustering to form 'risk corridors'; some factors (such as infrastructure investment) have a lag effect, while market speculation has an immediate impact.

• Resource Allocation Optimization: Pre-allocate resources to high-risk communities to improve return on investment
• Community Participation: Share prediction results to promote residents' participation in solution formulation
• Regulation and Planning: Implement measures such as inclusive zoning, community land trusts, anti-eviction protection, and small business support
• Policy Evaluation: Track the comparison between predictions and actual changes to optimize policy tools

• Data Quality: Develop cleaning and integration processes, establish a unified data warehouse, and cooperate with the government to ensure data updates
• Prediction Uncertainty: Use probabilistic prediction to provide confidence intervals
• Fairness: Conduct strict fairness audits to mitigate potential discriminatory impacts
• Interpretability: Use SHAP value analysis to explain the basis of predictions

Open-Source Value

Code, data (within privacy-permitted scope), and documentation are made public to enhance methodological transparency and transferability, and promote cross-domain collaboration.

Future Directions

• Real-time monitoring and early warning systems
• Policy simulation and impact assessment
• Multi-city comparative studies
• Integration of residents' subjective feelings and demands