Scholarlink

<p>Housing affordability is one of the signature governance challenges in New York City (NYC), with more than 52% of renter households cost-burdened and just under a third identified as severely cost-burdened, paying above 50% of gross income on housing. This study proposes a neighborhood-scale, explainable machine learning framework to predict severe housing cost burden at the Neighborhood Tabulation Area (NTA) level, drawing on a panel of 2,512 NTA-year observations spanning 239 NTAs from 2012–2022. Thirty modelling features derived from 49 raw socio-economic, housing market, and rent index variables are integrated from ACS 5-year estimates, eviction court records, and the Zillow Observed Rent Index (ZORI). Three gradient-boosted ensemble models—Random Forest, XGBoost, and LightGBM are benchmarked under a strict temporal train/validation/test split with 5-fold TimeSeriesSplit cross-validation to prevent data leakage. LightGBM achieved the highest predictive performance (Test <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(R^2\)</EquationSource> </InlineEquation>=0.9302, RMSE=0.03372, MAE=0.0185). SHAP TreeExplainer analysis identifies rent burden, vacancy rate, and rent-to-income ratio as dominant drivers, confirming that NYC’s rent escalation follows a systemic “burden escalator” across boroughs. Spatial diagnostics confirm no significant residual autocorrelation at both the borough level (Moran’s <i>I</i>=−0.0731, <i>p</i>=0.083) and NTA level (<i>I</i>=0.0204, <i>p</i>=0.238, <i>k</i>=5 kNN, <i>n</i>=239 NTAs), and ablation analysis confirms rental market variables drive the largest incremental contribution (<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\Delta R^2\)</EquationSource> </InlineEquation>=0.1629, 17.51%). A 8-instrument policy matrix is proposed, encompassing early-warning interventions, borough-differentiated zoning and voucher strategies, and anticipatory resource allocation. This framework demonstrates how explainable AI can operationalize predictive insights into actionable urban governance strategies, offering a replicable model for global cities facing housing crises.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

AI-Driven Housing Affordability Forecasting in New York City: An NTA-Level Panel Analysis using Ensemble Machine Learning

  • 期刊论文

  • 发表日期：

    2026年7月25日

查看全文
  • Azizur Rahman,
  • Nakib Uddin Ahmed,
  • Mehjabin Ferdous,
  • Kadirur Rahman Chowdhury

摘要

Housing affordability is one of the signature governance challenges in New York City (NYC), with more than 52% of renter households cost-burdened and just under a third identified as severely cost-burdened, paying above 50% of gross income on housing. This study proposes a neighborhood-scale, explainable machine learning framework to predict severe housing cost burden at the Neighborhood Tabulation Area (NTA) level, drawing on a panel of 2,512 NTA-year observations spanning 239 NTAs from 2012–2022. Thirty modelling features derived from 49 raw socio-economic, housing market, and rent index variables are integrated from ACS 5-year estimates, eviction court records, and the Zillow Observed Rent Index (ZORI). Three gradient-boosted ensemble models—Random Forest, XGBoost, and LightGBM are benchmarked under a strict temporal train/validation/test split with 5-fold TimeSeriesSplit cross-validation to prevent data leakage. LightGBM achieved the highest predictive performance (Test \(R^2\) =0.9302, RMSE=0.03372, MAE=0.0185). SHAP TreeExplainer analysis identifies rent burden, vacancy rate, and rent-to-income ratio as dominant drivers, confirming that NYC’s rent escalation follows a systemic “burden escalator” across boroughs. Spatial diagnostics confirm no significant residual autocorrelation at both the borough level (Moran’s I=−0.0731, p=0.083) and NTA level (I=0.0204, p=0.238, k=5 kNN, n=239 NTAs), and ablation analysis confirms rental market variables drive the largest incremental contribution ( \(\Delta R^2\) =0.1629, 17.51%). A 8-instrument policy matrix is proposed, encompassing early-warning interventions, borough-differentiated zoning and voucher strategies, and anticipatory resource allocation. This framework demonstrates how explainable AI can operationalize predictive insights into actionable urban governance strategies, offering a replicable model for global cities facing housing crises.