Predicting ash content and water content in coal using full infrared spectra and machine learning models

• Key findings: • Regression analysis using LassoCV, RidgeCV, ElasticNetCV, and LassoLarsCV revealed nonzero coefficients at specific wavenumbers, indicating the impact of IR intensities on predicting ash content and water content in coal samples. • The identified wavenumbers with nonzero coefficients were correlated with distinct spectral features observed in the IR spectra of the coal samples, such as peaks, bands, and vibrational frequencies. • These spectral features at specific wavenumbers enabled precise predictions of ash content and water content in the coal samples, as demonstrated by strong correlations between the predicted and experimental values. • The findings contribute to the simple non-destructive determination of coal ash content and water content across various applications. • These research findings highlight the potential of machine learning models in predicting ash content and water content in coal samples using infrared spectra. The aim of this study was to predict ash and water contents in coal samples using machine learning regression techniques, specifically LassoCV, RidgeCV, ElasticNetCV and LassoLarsCV. The analysis focused on finding non-zero coefficients at specific wavenumbers and highlighted the influence of infrared (IR) intensity on prediction accuracy. These determined wavenumbers were correlated with experimental ash and water contents in coal samples. The study showed a strong relationship between spectral features and regression coefficients, thus enabling accurate prediction of ash and water contents. For ash content, significant spectral features were identified at around 600 cm⁻¹ and 1600 cm⁻¹, corresponding to C=C and aromatic carbon vibrations. The prediction of water content was significantly influenced by O-H vibration at around 3700 cm⁻¹. The performance of the regression models was evaluated by comparing the predicted ash and water contents with experimental data, thus ensuring a strong correlation between the predicted and experimental values. This study highlighted the effectiveness of regression analysis and machine learning models in predicting coal properties and provided valuable information for better assessment of direct coal parameters.