Design of an Integrated Model Using XGBoost, ConvLSTM, and Multiple Agent DQN for Spatio-Temporal AQI prediction for Healthcare Enhancements

Abstract

AQI can be defined as the monitoring of air quality index, which has turned out to be an essential need for people living in large cities and industrial areas, as it seeks to lessen adverse health effects. However, most current models in regard to AQI prediction & healthcare enhancements barely take into account complex spatio-temporal dynamics of the pollutant and meteorological factors. Most of the existing methods face a trade-off among the prediction accuracy, computation efficiency, and adaptability across diverse environmental conditions. It proposes a multivariate AQI prediction & healthcare enhancements model through advanced ensemble machine learning and deep learning methods in the perspective of geographically diversified urban and industrial areas of Delhi, India Geographies. In this regard, an XGBoost integrated with RFE toward AQI prediction & healthcare enhancements has been proposed with optimized key parameters: PM2.5, PM10, NO2, CO, SO2, temperature, humidity, wind speed, and aerosol parameters. The model provides high accuracy with computational efficiency. Further, the ConvLSTM combined with Kriging enhances spatial and temporal prediction capabilities in filling gaps in the monitoring data from residential, industrial, and heavy-traffic areas like R.K. Puram, Wazipur, and ITO. Spatial interpolation by using Kriging will ensure complete coverage at places where monitoring stations are not available. This makes real-time optimization utilize the Multiple Agent DQN to propose dynamic interventions for mitigating the level of pollution-particularly, the traffic and industrial emission. DTW with DBSCAN finally emphasized the clustering of pollution that helps identify high-risk areas like Anand Vihar and Okhla. The proposed integrated approach significantly improves AQI prediction & healthcare enhancements with actionable insights for policymakers and environmental regulators. Key results: A better performance in the AQI prediction & healthcare enhancements with an average absolute error of ~3.5 units and a reduction in episodes of high pollution by 15%. These gains are immediately applicable to urban air quality management with tangible public health benefits across a wide range of urban sectors.