Agricultural productivity in developing nations like India faces escalating threats from air pollution, yet conventional yield prediction models often neglect this critical environmental stressor. This study bridges this gap by integrating air pollution metrics (PM2.5, SO₂, NOx, SPM, RSPM) with soil, climatic, and crop data to develop machine learning (ML)-based tools for yield prediction and crop recommendation. Unlike prior studies, this work uniquely integrates air pollution metrics into agricultural Machine learning models. We propose a dual framework: (1) a regression task predicting yields of five major crops (rice, wheat, maize, barley, jute) using Random Forest, XGBoost, SVM, and Ridge Regression, and (2) a classification task recommending optimal crops under pollution stress via Naïve Bayes, SVM, XGBoost, and Logistic Regression. Leveraging datasets spanning air quality, soil nutrients, and historical yields, our results demonstrate that XGBoost achieves superior performance in yield prediction (R2 = 0.96 for wheat), while Random Forest excels in crop recommendation (test accuracy = 91.5%). Analysis reveals pollutants like SO₂ and NO₂ significantly reduce yields, with data experiments confirming model robustness to pollution variability. By incorporating air quality indicators into agricultural decision-making, this work advances precision farming strategies, offering actionable insights for farmers and policymakers to mitigate pollution-driven losses and enhance sustainability in agro-climatically vulnerable regions.