Improving the short-term prediction of dissolved carbon monoxide using a combination of Light GBM and meta-heuristic algorithms

Abstract

In this study, the prediction of carbon monoxide pollutants on a short-term scale has been investigated according to some input data sources, comprising gas concentrations related to air quality and weather features. Utilizing a hybrid modeling approach that integrates the Light Gradient Boosting Machine with several meta-heuristic optimization algorithms such as Chaos Game Optimization, Aquila Optimizer, and others, we aimed to optimize the hyperparameters of the Light GBM to enhance predictive accuracy. The application of a K-fold cross-validation technique with K=5 helped in preventing overfitting. By conducting a case study on a real dataset collected from a gas multi-sensor device, it was found that the hybrid model combining the Light Gradient Boosting Machine with Chaos Game Optimization demonstrated superior performance compared to other models. The values of the coefficient of determination, Root Mean Square Error, Mean Absolute Error, and Mean Absolute Percentage Error for this model based on test data are 0.99, 0.0393, 0.0301, and 0.0052, respectively. These results underscore the effectiveness of the hybridization approach in providing highly accurate predictions for short-term carbon monoxide concentrations, offering a valuable tool for environmental monitoring and enhancing public health safeguards.