Explore the vast range of titles available.

A hybrid AQI time series prediction model is proposed based on EWT-SE-VMD secondary decomposition, ICA ( imperialist competitive algorithm ) feature selection, and ESN ( echo state network ) neural network. Firstly, EWT ( empirical wavelet transform ) and VMD ( variational mode decomposition ) are used to decompose the original AQI time series into several stable and reliable subseries. Then, the ICA is used to select features of the above subseries for the ESN prediction model. Finally, the optimized feature variables are put into the ESN deep network to establish a prediction model of each AQI subseries and obtain the future AQI index. According to the experimental results of the daily AQI series in Beijing, Tianjin, and Shijiazhuang, we find that (a) among all decomposition methods, the proposed secondary decomposition method (EWT-SE-VMD) performs best in processing data; (b) it is proved that the proposed hybrid model has broad application prospect and research value in the AQI prediction field.

The monitoring of air pollution through the air quality index (AQI) is a fundamental tool in ensuring public health protection. Accurate prediction of air quality is necessary for the timely implementation of measures to control and manage air pollution, thereby mitigating its detrimental impact on human health. A novel hybrid prediction model is proposed, which is EMD-KMC-EC-SSA-VMD-LSTM. Raw AQI index data are decomposed into intrinsic mode functions (IMFs) by empirical mode decomposition (EMD) method. Subsequently, sample entropy (SE) is utilized to assess the intricacy of IMFs, and K-means clustering (KMC) is used to reconstruct them into joint intrinsic mode functions (Co-IMFs). Then, the variational mode decomposition (VMD) is used to transform the complex Co-IMF0 into simpler IMFs. Long short-term memory (LSTM), optimized either by the Sparrow Search Algorithm (SSA), is applied to forecast all IMFs, generating the first prediction sequence. To further refine the forecasting, an error correction (EC) technique is adopted. The error sequence is obtained by subtracting the forecasting sequence from the raw sequence, which is then decomposed by EMD-SSA-VMD. Subsequently, SSA-LSTM is engaged to forecast the decomposed error sequence, generating the error forecasting sequence. Finally, the forecast outcomes are combined with the error predictions to generate the final AQI prediction sequence. The proposed approach undergoes validation across four urban centers and undergoes comparison against a set of eight prediction models. Experimental findings underscore the heightened precision of this hybrid forecasting model in predicting AQI metrics.

The air quality index (AQI) prediction is important to evaluate the effects of air pollutants on human health. The airborne pollutants have been a major threat in Delhi both in the past and coming years. The air quality index is a figure, based on the cumulative effect of major air pollutant concentrations, used by Government agencies, for air quality assessment. Thus, the main aim of the present study is to predict the daily AQI one year in advance through three different neural network models (FF-NN, CF-NN and LR-NN) for the year 2020 and compare them. The models were trained using AQI values of previous year (2019). In addition to main air pollutants like PM 10 /PM 2.5 , O 3 , SO 2 , NOx, CO and NH 3 , the non-criteria pollutants and meteorological data were also included as input parameter in this study. The model performances were assessed using statistical analysis. The key air pollutants contributing to high level of daily AQI were found to be PM 2.5 /PM 10 , CO and NO 2 . The root mean square error (RMSE) values of 31.86 and 28.03 were obtained for the FF-NN and CF-NN models respectively whereas the LR-NN model has the minimum RMSE value of 26.79. LR-NN algorithm predicted the AQI values very closely to the actual values in almost all the seasons of the year. The LR-NN performance was also found to be the best in post-monsoon season i.e., October and November (maximum R 2  = 0.94) with respect to other seasons. The study would aid air pollution control authorities to predict AQI more precisely and adopt suitable pollution control measures. Further research studies are recommended to compare the performance of LR-NN model with statistical, numerical and computational models for accurate air quality assessment.

The air quality index (AQI) forecast in big cities is an exciting study area in smart cities and healthcare on the Internet of Things. In recent years, a large number of empirical, academic, and review papers using machine learning (ML) for air quality analysis have been published. However, most of those studies focused on traditional centralized processing on a single machine, and there had been few surveys of federated learning (FL) in this field. This overview aims to fill this gap and provide newcomers with a broader perspective to inform future research on this topic, especially for the multi-model approach. In this survey, we went over the works that previous scholars have conducted in AQI forecast both in traditional ML approaches and FL mechanisms. Our objective is to comprehend previous research on AQI prediction including methods, models, data sources, achievements, challenges, and solutions applied in the past. We also convey a new path of using multi-model FL, which has piqued the computer science community’s interest recently.

Two data-driven algorithms, back propagation neural network (BPNN) and support vector regression (SVR), are adopted to predict air quality index (AQI) in Jiangsu Province. Meanwhile, the static model, the dynamic model of daily training and the half-daily training are established to validate the performance of algorithms comprehensively. The fundamental advantage of support vector is that less data in the full set is selected to efficiently capture the whole characteristics, whereas BPNN is more accurate in description of the high dimensional models since the parameters are well trained. The comparisons between two algorithms for the above models demonstrate that BPNN outperforms SVR in terms of accuracy since most of the mean absolute percentage errors of BPNN are less than 10%, which decrease 3% compared with that of SVR, whereas the computational cost of SVR is much less than that of BPNN. Furthermore, a novel hybrid model, the SVR-BPNN model, is proposed to further predict and analyze the AQI, which performs as fairly well as BPNN but is less time-consuming.

Globally, there are significant worries about the rise in air pollution (AP) from substances that are harmful to human health, different living forms, and unfavorable environmental imbalances. To overcome the problem, AI-based prediction model is the need of the hour. Therefore, an attempt was made to develop a novel AP prediction system based on Xavier Reptile Switan-h-based Long-Short Term Memory (XRSTH-LSTM), which undergoes fine-tuning at various steps such as pre-processing, attribute extraction, and air-quality index prediction, in order to reduce computational cost and also to increase accuracy as well as precision. The dataset used to train the proposed methodology is Air Quality Data in India (2015–2020), taken from publically available sources Kaggle. The dataset includes information on the AQI and air quality at different stations in numerous Indian cities at hourly and daily intervals. The accuracy has been calculated using MSE, MAPE, RMSE, precision, recall, and F-measure. The robustness of the proposed model is tested using parameters such as negative predicted value and Mathew correlation coefficient. The proposed model is found to efficiently process air quality with an improved accuracy of 98.52% and precision of 99.79%, which is 0.74% higher than the existing state-of-the-art model. The testing findings showed that the proposed approach worked better than the current models and offered a higher rate of accuracy in predicting air pollution.

From a macro-perspective, based on machine learning and data-driven approach, this paper utilizes multi-featured data from 31 provinces and regions in China to build a Bayesian network (BN) analysis model for predicting air quality index and warning the air pollution risk at the city level. Further, a two-layer BN for analyzing influencing factors of various air pollutants is developed. Subsequently, the model is applied to forecast the trends of temporal and spatial changes in the form of probabilistic inference and to investigate the degree of impact incurred from individual influencing factors. From the comparisons with the results obtained from other machine learning approaches and algorithms such as neural networks, it is concluded that by comprehensively using the established BN, one can not only reach a monitoring and early warning accuracy rate of 90% but also scrutinize and diagnose the main cause of air pollution risk changes from the perspective of probability.

A prompt and accurate prediction of air quality index (AQI) has become a necessity to tackle the mounting environmental threats. This paper proposes a feature-driven hybrid method for hourly, 3-step-ahead, and deterministic AQI prediction, which includes three modules. In Module 1, an “extract-merge-filter” procedure of feature engineering is created to capture the potential features from the AQI series. Ten feature sets are generated as candidates. In Module 2, six models including Light Gradient Boosting Machine, Extreme Gradient Boosting, Long Short-Term Memory, Convolutional Neural Network, Multilayer Perceptron, and Deep Neural Network are developed as base predictors and performed on the candidate features. In Module 3, predictors are first matched with their optimal features using a comprehensive metric, and then combined in an optimized ensemble using OPTUNA. A case study on the AQI data from four different Chinese cities is carried out to demonstrate the method. The experimental results show the following: (1) Feature engineering significantly boosts prediction performance and provides interpretable findings for practical use. (2) Customized input of features to the predictors is more effective than a fixed input and can rise the performance to a higher level. (3) OPTUNA is a promising tool for optimizing ensemble weights. The final ensemble model is superior to single machine learning models and has a good robustness.