Machine learning optimized LIBS spectral feature extraction for indoor carbon concentration dynamics and forecasting

Abstract

Domestic energy use and indoor human activities generate substantial carbon emissions, making accurate monitoring of indoor carbon concentrations essential for identifying emission sources and assessing associated exposure risks. Laser-induced breakdown spectroscopy (LIBS) enables rapid, multi-element, and in situ detection of aerosols, exhibiting excellent sensitivity to carbon-related spectral characteristics. Combining LIBS with machine learning allows accurate prediction of carbon content by leveraging complex spectral patterns, with systematic hyperparameter optimization using the Sparrow Search Algorithm (SSA) and interpretable insights from SHapley Additive exPlanations (SHAP) analysis. Regression models, including Random Forest (RF), support vector regression (SVR), and least squares support vector machine (LSSVM), were applied to quantify spectral contributions and reveal the relationships between spectral features and carbon concentration. The models achieved high predictive accuracy, with R² values of 0.976 (RF), 0.979 (SVR), and 0.980 (LSSVM) for candles, and 0.943 (RF), 0.952 (SVR), and 0.980 (LSSVM) for fruit charcoal, demonstrating their overall effectiveness in carbon content analysis, with LSSVM showing slightly superior performance. To capture temporal spectral dynamics, a dynamic time-series prediction model was constructed by coupling an attention mechanism with a bidirectional long short-term memory (Bi-LSTM) network. This model successfully forecasts carbon concentration trends by learning long-range dependencies and salient temporal features, achieving R² values of 0.929 for candles and 0.882 for fruit charcoal. Overall, the integrated LIBS–machine learning approach enables fast, accurate, and interpretable carbon detection, substantially enhancing the precision of carbon source attribution and providing a scientific basis for indoor air quality monitoring and emission source management.