Comparative Analysis of Machine Learning Approaches for Urban Heat Island Prediction: Random Forest, XGBoost, and Convolutional Neural Networks (CNN)

Keywords: urban heat, New York City, census block, remote sensing, NDVI, Landsat, machine learning, Random Forest, urban planning, climate resilience
Abstract Type: Poster Abstract

Abstract

This study examines the performance of different machine learning models—Random Forest, XGBoost, and Convolutional Neural Networks (CNN)—in predicting urban heat intensity across New York City. Using high-resolution remote sensing data, including NDVI, digital elevation models (DEM), and land cover classifications, this research assesses the influence of these geographic features on urban temperature distribution. The study employs a spatial cross-validation approach, training models on one half of the city and predicting temperatures for the other half. Model performance is evaluated using metrics such as Mean Absolute Error (MAE) and R² score. The results highlight key factors contributing to urban heat and compare the effectiveness of different learning algorithms in capturing spatial temperature variations. These findings provide insights for urban planners in mitigating heat stress through data-driven strategies.