Particulate matter (PM) is a complex and multidimensional aspect of atmospheric chemistry that is represented by a variety of small solid and liquid droplets suspended in the atmosphere. These minuscule particles come in a variety of sizes and compositions, with dimensions ranging from a few nanometers to a few tens of micrometers. The causes of PM are closely linked to the environment on Earth, resulting from a complicated interaction between natural and human-caused factors (Ogrizek et al., 2022). A thorough investigation of the chemical components and generation mechanisms of PM is necessary to comprehend its complex character. PM originates from a variety of sources, including particles produced by the Earth’s natural processes (Rai, 2016). These include mineral dust particles created by the erosion of rocks and soil, as well as biological elements like pollen and spores from plants and microbes (Aarnink et al., 2010). Sea salt aerosols are produced by the breaking waves in the ocean and add to the natural PM concentration. These naturally occurring particulates are crucial to the functioning of Earth’s ecosystems because they help in transferring nutrients, generate clouds, and serve as the building blocks for the creation of raindrops (Ubaid et al., 2019). Human activity significantly contributes to the atmospheric PM, releasing a variety of particles from combustion, construction, and industrial processes. These sources often contain heavy metals and organic chemicals, resulting in a dynamic and ever-changing composition (Chen et al., 2015).
The increase in PM pollution poses an increasing threat to the environment. These microscopic particles can have profound and far-reaching effects on human health, ecosystems, and the environment. Owing to the fact that PM can enter the respiratory system deeply and cause a variety of respiratory and cardiovascular disorders, it poses a serious risk to human health (Hamanaka & Mutlu, 2018). In addition, PM harms the environment due to its participation in the production of smog, decreased visibility, and deposition of natural ecosystems, which lowers the quality of the soil and water (Kim et al., 2016). Therefore, the need for effective addressing of PM pollution necessitates the use of intricate and advanced models to comprehend its origins, distribution, and ecological impacts (G et al., 2013).
Before the advancement of advanced machine learning (ML) and artificial intelligence (AI) techniques, managing the intricate issues related to particulate matter pollution was difficult. Policymakers and environmental scientists mainly used manual data analysis and simple equations in their older models (Liu et al., 2018). The complex dynamics of PM sources and dispersion patterns were difficult for these older methods to capture. Furthermore, they could not properly utilize the potential of the enormous datasets that are already accessible (Luo et al., 2019). Consequently, the shortcomings of these conventional approaches limited the ability to minimize the harmful impacts of PM pollution and optimize environmental management (Wu et al., 2018).
For decades classical methods of managing air pollution have been associated with several uncertainties and approximations owing to their complex and nonlinear nature(Costache et al., 2019; Mohammadi et al., 2020; Pham et al., 2019; Sammen et al., 2021). It is well known based on the developed traditional techniques of controlling PM 2.5 that utilizing AI models for PM 2.5 prediction offers substantial advantages by significantly enhancing the accuracy and efficiency of air quality forecasting. Artificial intelligence (AI) algorithms improve the overall understanding of PM 2.5 modelling, enabling them to adapt to complex environmental data and ensuring real-time, globally applicable insights. This approach not only accelerates computational processes but also supports informed decision-making for effective and sustainable air quality management. However, coupling nature-inspired meta-heuristic optimization algorithms and ensemble machine learning (ML) presents a novel approach to modelling particulate matter (PM 2.5) concentrations, which are critical to air quality management. By mimicking evolutionary processes, these algorithms can efficiently navigate complex search spaces to calibrate ensemble ML models, thereby enhancing their predictive accuracy(Baig et al., 2023). This serves as the fundamental in creating sustainable, eco-friendly strategies for air quality control, leveraging the inherent adaptability of nature-inspired algorithms to optimize the weights and hyperparameters of ML models (Uniyal et al., 2022).
Recently, PM 2.5 modelling using advanced AI and spatial mapping attracted a huge researcher’s attention. For instance, Tai et al. (Tai et al., 2010) explored correlation-based and ML models for the prediction of PM 2.5 using metrological parameters over 11 years of US data and showed that weather conditions explain about 50% of PM2.5 variations. In 2012, Tai et al. (Tai et al., 2012) employed several models to predict PM 2.5 using metrological variables for the Impact of 2000–2050 climate change. In 2018, Polezar et al. (Polezer et al., 2018) employed nonlinear artificial neural network (ANN), Multilayer Perceptron (MLP), Extreme Learning Machines (ELM) and Echo State Networks (ESN) to evaluate the impact of PM 2.5 on human health. The outcomes indicated that ANN is a more reliable method for a smaller amount of data. In addition, Wei et al. (Wei et al., 2020) in 2020 utilize computational randomized trees to estimate PM 2.5 across China based on 1km resolution. A more recent study in 2023 was conducted by Gokul et al. (Gokul et al., 2023) in Hyderabad city India in which PM 2.5 was predicted using spatial method, single and deep learning approaches. The outcomes indicated the prediction skills of LSTM (Long Short-Term Memory) over XGBoost. The other that attempt to predict PM 2.5 in several region across the globe include (Hu et al., 2014; Lary et al., 2016; Liang et al., 2018; Lin et al., 2015; Meng et al., 2021; Pandya et al., 2023; Prunicki et al., 2018; Xue et al., 2019; Zhang et al., 2021).
It can be seen that PM 2.5 received significant attention using single AI model, however it is very crucial to optimize the standalone model using optimization techniques(Moayedi et al., 2023; Simon, 2008; Yaseen et al., 2018). Optimization is essential in prediction as it enhances model accuracy, ensures computational efficiency, and improves generalization capabilities, which are crucial for effective and cost-efficient decision-making, particularly in environmental monitoring like PM 2.5 assessment. Moreover, the application of these advanced techniques to PM 2.5 modelling is a testament to the potential of AI in environmental science. By harnessing the stochastic yet structured search mechanisms of meta-heuristic algorithms, such as Ant Colony Optimization (ACO) or Simulated Annealing (SA), the ensemble models can simulate and predict PM 2.5 variations with high precision. This allows for real-time monitoring and proactive policymaking, mapping a course towards a sustainable future with cleaner air, thus implementing both environmental and public health imperatives(Nabipour et al., 2020).
Similarly, the motivation for the study is to improve the monitoring and modelling of PM 2.5, a significant air pollutant with serious health implications, due to its capacity to access deep into the respiratory system. The study aims to enhance the accuracy of PM 2.5 predictions, which are vital for developing effective air quality management strategies to protect public health and ensure a sustainable environment. It is worth mentioning based on the above literature that the gap identified in the study is the need for more efficient and accurate prediction models for PM 2.5 concentrations. Existing models may not fully capture the complexity of air quality data or might not be optimized for the best performance. To address this, the study integrates nature-inspired meta-heuristic optimization algorithms with ensemble machine learning techniques to improve the prediction models. The use of specific algorithms like ANN-PSO (Artificial Neural Network with Particle Swarm Optimization) is highlighted as particularly effective, outperforming other techniques in both calibration and validation phases. The research also explores the potential of the Neural Network Ensemble (NNE) over simpler averaging (SA) ensemble methods, indicating a methodological improvement in the field of air quality modelling.