Feature selection (FS) constitutes a critical stage within the realms of machine learning and data mining, with the objective of eliminating irrelevant features while guaranteeing model accuracy. Nevertheless, in datasets featuring a multitude of features, choosing the optimal feature poses a significant challenge. This study presents an enhanced Sand Cat Swarm Optimization algorithm (MSCSO) to improve the feature selection process, augmenting the algorithm’s global search capacity and convergence rate via multiple innovative strategies. Specifically, this study devised logistic chaotic mapping and lens imaging reverse learning approaches for population initialization to enhance population diversity; balanced global exploration and local development capabilities through nonlinear parameter processing; and introduced a Weibull flight strategy and triangular parade strategy to optimize individual position updates. Additionally, the Gaussian–Cauchy mutation strategy was employed to improve the algorithm’s ability to overcome local optima. The experimental results demonstrate that MSCSO performs well on 65.2% of the test functions in the CEC2005 benchmark test; on the 15 datasets of UCI, MSCSO achieved the best average fitness in 93.3% of the datasets and achieved the fewest feature selections in 86.7% of the datasets while attaining the best average accuracy across 100% of the datasets, significantly outperforming other comparative algorithms.

Accurate prediction of PM_2.5 concentration is important for pollution control, public health, and ecological protection. However, due to the nonlinear nature of PM_2.5 data, the accuracy of existing methods suffers and performs poorly in both short-term and long-term predictions. In this study, a deep learning hybrid prediction model based on clustering and quadratic decomposition is proposed. The model utilizes the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to decompose the PM_2.5 sequences into multiple intrinsic modal function components (IMFs), and clusters and re-fuses the subsequences with similar complexity by permutation entropy (PE) and K-means clustering. For the fused high-frequency sequences, a secondary decomposition is performed using the whale optimization algorithm (WOA) optimized variational modal decomposition (VMD). Finally, the nonlinear and temporal features are captured for prediction using the long- and short-term memory neural network (LSTM). Experiments show that this proposed model exhibits good stability and generalization ability. It does not only make accurate predictions in the short term, but also captures the trends in the long-term prediction. There is a significant performance improvement over the baseline models. Further comparisons with existing models outperform the current state-of-the-art models.

PM_2.5 pollution poses an important threat to the atmospheric environment and human health. To precisely forecast PM_2.5 concentration, this study presents an innovative combined model: EMD-SE-GWO-VMD-ZCR-CNN-LSTM. First, empirical mode decomposition (EMD) is used to decompose PM_2.5, and sample entropy (SE) is used to assess the subsequence complexity. Secondly, the hyperparameters of variational mode decomposition (VMD) are optimized by Gray Wolf Optimization (GWO) algorithm, and the complex subsequences are decomposed twice. Next, the sequences are divided into high-frequency and low-frequency parts by using the zero crossing rate (ZCR); the high-frequency sequences are predicted by a convolutional neural network (CNN), and the low-frequency sequences are predicted by a long short-term memory network (LSTM). Finally, the predicted values of the high-frequency and low-frequency sequences are reconstructed to obtain the final results. The experiment was conducted based on the data of 1009A, 1010A, and 1011A from three air quality monitoring stations in the Beijing area. The results indicate that the R² value of the designed model increased by 2.63%, 0.59%, and 1.88% on average in the three air quality monitoring stations, respectively, compared with the other single model and the mixed model, which verified the significant advantages of the proposed model.