The combustion of fuels in industries is one of the sources of greenhouse gases, posing a considerable threat to human health and the natural environment. The monitoring of the combustion process and the measurement of combustion degrees, therefore, are crucial and significant. In this work, an experimental system was developed based on Laser-induced breakdown spectroscopy (LIBS). Charcoals with different combustion degrees were taken as samples, and in the LIBS spectra of charcoals, the characteristic line of C and the molecular bands of CN and CaO were observed. Moreover, a univariate calibration model based on the exponential fitting curve was constructed to predict the combustion degree according to the variation of the C I line intensity under different combustion degrees. Furthermore, different algorithms including support vector machine (SVM), partial least squares (PLS), random forest (RF), and convolutional neural network (CNN) were applied to establish the multivariate calibration model, and the prediction performance was compared according to the 10-fold cross-validation. Particle swarm optimization (PSO) was used to optimize hyperparameters of the selected SVM-based multivariate calibration model. Principal component analysis (PCA) and linear discriminant analysis (LDA) were employed to extract spectral features and reduce dimensions for the improvement of the model's performance, with the final root mean square error of prediction (RMSEP) and mean relative error (MRE) of the LDA-PSO-SVM calibration model of 0.035 and 3.877%, respectively. Finally, c. The results prove feasible to realize the real-time determination of combustion degree to control carbon emissions and monitor carbon concentration through the LIBS-based method with machine learning, which also provides new ideas for the application of LIBS in the atmospheric field.