Hydrogenated amorphous carbon (a-C:H) thin films synthesized by plasma-enhanced chemical vapor deposition (PECVD)are widely used in various fields of science and technology. For example, they are essential as etching hard masks for the production of 3D-NAND flash memory, because of their high selectivity to Si-based materials and chemical stability. With the demand for higher aspect ratio etching, there is a strong need to improve etching resistance and reduce residual stress of a-C:H masks. They are also widely used in the field of tribology as an important coating for improving friction resistance and slidability. Analyzing the plasma process in which synergistic effects among reactive species and higher-order reactions contribute is generally difficult. Therefore, similar to other process technologies, machine learning-based condition optimization approaches have been actively investigated in recent years. However, machine learning models are generally black boxes, and it is difficult to understand their mechanism of action. Also, the versatility of the learning model is low between different devices. On the other hand, reactive species measurement is effective for understanding the reaction mechanism of the plasma process, but the synergy of multiple reactive species is still extremely complicated. In this study, the contribution of radical species to the etching resistance of a-C:H thin films were quantitatively analyzed using machine learning.The deposition of a-C:H by PECVD using H2/C3H6/CH4 plasma and etching with O2 plasma were alternately repeated at 120 conditions. The radicals generated during the film deposition were measured by a quadrupole mass spectrometer (QMS), and film thickness change during the etching was measured by spectroscopic ellipsometry (SE). A random forest model was learned to predict the etching rate from the mass spectra, and the contribution of each radical to the etching rate was quantitatively evaluated by SHapley additive exPlanation (SHAP).The distribution of SHAP values for the 10 most influential radicals is shown in Figure 1. The positive and negative SHAP values represent the correlation between QMS intensity and SHAP value. The radicals with high hydrogen ratios such as CH~CH4, C2H5, C4H9, C5H9, and C5H11 contributed to the increase of etching rate, while the radicals with low hydrogen ratios such as C3H3 and C5H5 contributed to the decrease of etching rate. The deposition of radicals with low hydrogen ratios and multiple bonds, and the effect of hydrogen extraction in the film are considered to have increased the density of a-C:H and reduced the etching rate. It was found that the control of C/H ratio of radicals as film precursors is important to improve etching resistance. Comparisons with different data set range or other contribution degrees such as LIME enable us to understand the mechanism of action of local active species. By using the contribution analysis methods, it is possible to quantitatively analyze the synergistic effect of reactive species in the process plasma. Figure 1