Spatial prediction and visualization of PM _2.5 susceptibility using machine learning optimization in a virtual reality environment

Particulate Matter (PM2.5)-based air pollution is a severe menace to health and the environment worldwide. Traditional methods for predicting and mapping PM2.5 are usually lagging due to the complexity and nonlinearity between different participating parameters. Moreover, there is also a significant gap in the literature regarding visualizing air pollution data using immersive technologies such as Virtual Reality (VR). This paper overcomes these shortcomings by combining state-of-the-art machine learning advancements with new visualization techniques. It improves the spatially predicted accuracy of PM2.5 using Support Vector Regression (SVR) optimized with Particle Swarm Optimization (PSO). Moreover, the visualization is depicted through the VR-based application in an immersive way to comprehend PM2.5 risk maps. To this end, 13 spatial factors influencing PM2.5, including topographic, climatic, land use, and population density factors in Tehran, Iran, were considered for modeling and spatial visualization. The susceptibility maps of PM2.5 generated by the hybrid SVR-PSO model demonstrated an accuracy of 96.3% using the Receiver Operating Characteristic (ROC) curve. The evaluation results of the VR systems from the Virtual Reality Neuroscience Questionnaire (VRNQ) and System Usability Scale (SUS) for spatial visualization showed that they had high graphics capabilities and equipment for the spatial prediction of PM2.5.