AIR QUALITY PREDICTION USING DEEP LEARNING MODELS: DETAILED OVERVIEW

Abstract

The goal of a real-world application of air quality forecasting systems is to provide a real-time platform for monitoring air quality that is user-friendly. A Bidirectional Stacked LSTM and Single-Step ANN were integrated into an interactive web-based application using Streamlit to create real-time air quality forecasting systems that will give access to air quality forecasts to government agencies, public health authorities, and the public. Using real-time data such as PM2.5, PM10, NO₂, CO, temperature, and humidity, users can obtain both single-step and multi-step AQI forecasts. The model forecasts are updated in real-time and shown, which gives users immediate feedback to inform them for short-term activities like health advisories and long-term like pollution action plans. Assessment of the application included accuracy, efficiency, and ease of use that will provide timely and actionable forecasts. The system was built to have high scalability and can support high data and user capacity to allow for use in the cloud. There are some hurdles that require attention to make real-time air quality forecasting systems practical such as data quality, model interpretability, and computational cost. However, it is clear that the adaptation of deep learning models in real-time air quality forecasting systems, has the potential for improving public health management and environmental policy decisions in practice.