Air pollution in cities is a constant threat to the health of the public and the environment. Fine particulate matter (PM2.5) is especially bad for your health because it is strongly linked to heart and lung diseases. The World Health Organization and other international health organizations say that air pollution is one of the biggest environmental risks in the world. To move from reactive monitoring to anticipatory management, we need strong predictive analytics that can model nonlinear atmospheric dynamics. This research formulates and evaluates statistical and machine learning methodologies for short-term PM2.5 forecasting, utilizing a simulated multi-year urban dataset that emulates authentic meteorological and emissions dynamics. We use out-of-sample performance metrics to compare the Multiple Linear Regression (MLR), ARIMA, Random Forest, and Long Short-Term Memory (LSTM) models. Results show that nonlinear ensemble and deep learning methods work much better than traditional regression models, especially when pollution levels are very high. Nonetheless, interpretable statistical models continue to be significant for policy formulation and regulatory clarity. The results back the idea of adding predictive modelling to sustainability governance frameworks that are in line with the UN's Sustainable Development Goals, especially those that have to do with health and sustainable cities.