The environment has been greatly damaged by the quickening pace of urbanization, underscoring the need for better management of pollution and climate change indicators. By creating smart city ecosystems, the Fourth Industrial Revolution (4IR) offers the opportunity to address these issues. These ecosystems combine governance, environmental monitoring, and astute planning to produce safe and sustainable urban settings. This research explores how machine learning models can be used to forecast air quality, providing data-driven and environmentally conscious alternatives for smart cities. Using two deep learning methods (MLP regressor and LSTM) and four machine learning methods (AdaBoost, SVR, RF, and KNN), we provide a beginning-to-end air quality prediction model designed for smart city applications. All four major cities in India—Delhi, Kanpur, Lucknow, and Jalandhar—are the subject of the study. It makes use of meteorological data, such as wind direction and speed, and air quality data, for pollutants like PM2.5, PM10, for example, O3, and CO. Understanding the connection between correlation and variances of the variable in question in air pollution prediction is the main objective. Feature optimization, which lowers dimensionality and eliminates superfluous features, is used to improve PM2.5 prediction accuracy. The findings show that LSTM models, which achieve high R2 values of 0.990 in Delhi 0.995 in Kanpur 0.9 in Lucknow, and 0.903 in Jalandhar, outperform other models in forecasting PM10 and PM2.5 concentrations. The study determines that the most important variables to monitor are PM2.5, PM10, NO2, the speed of the wind, and humidity. The process of feature optimization highlights the important aspects influencing air quality and improves the interpretability of the model. These results offer policymakers insightful information that will help them create practical plans to reduce air pollution and encourage sustainable urban life in smart cities.