A multi-stage hybrid optimization framework based on Crocodile Optimization and Beetle Antennae Search for high-precision air quality analytics

Rapid industrialization and urbanization continue to impair global air pollution, making accurate air quality forecasting critical for effective environmental management. The main aim of this paper is to create a hybrid intelligent system that combines improved feature selection optimization to accurately predict applications. A collection of computerized weather stations with the main chemicals, such as CO and hydrogen, is used in the structure. The Crocodile Optimization, Beetle Antenna Search, and Pathfinder Algorithm (CSO–BFA) present superior predictive performance. The bio-inspired methods work concurrently to enhance the feature selection based on learning the most important variable quantity in complex spatial data while filtering out noise and dimensionality. The optimized feature collection is integrated into a strong prediction model for accurate prediction. The performance of the training process was examined by utilizing an improved algorithm that implemented a Multilayer Perceptron Backpropagation Neural Network (MPBPNN) to train and predict air pollution. Based on the practical results, it was determined that the hybrid model showed improved functionality in conditions of classification and prediction. The model's CSO–BFA–MPBPNN framework achieves the highest (0.9029) and the lowest error rates (RMSE = 0.4167, MAE = 0.0976), and the prediction accuracy is 94.05%. In this model, the multiple optimization techniques are integrated across several strategies, resulting in high accuracy. The system delivers a reliable, real‑time tool for early warning and active environmental management. Air pollution-related health and environmental risks can be minimized with informed decision-making.