Coal quality parameters, such as carbon content and ash yield, significantly influence the economic and metallurgical value of coal. Accurate determination of these parameters is crucial for optimizing industrial processes as they affect energy efficiency, processing costs, and pollution emissions. The standard ultimate and proximate analyses are time-consuming and resource-intensive. To offer an efficient alternative to the exhaustive standard methods, the present study attempted the use of mid-infrared Fourier Transform Infrared spectroscopy (FTIR) combined with advanced machine learning models to quickly and accurately predict coal carbon content and ash yield. Eighteen coal samples from the Johilla coalfield in Madhya Pradesh, India, was analyzed using FTIR spectroscopy across the range of 4000 to 350 cm−1. Machine learning algorithms like partial least squares regression (PLSR), random forest regression (RFR), support vector regression (SVR), and a multimodel estimation (MME) approach utilizing the average of the three models were employed to predict carbon content and ash yield. The MME model delivered enhanced precision and robustness compared to the individual models, with an R2 of 0.962, RMSE (%) of 22.812, and MBE (%) of 4.042 for carbon content and an R2 of 0.841, RMSE (%) of 37.081, and MBE (%) of 8.039 for ash content.