In this study, we developed a data-driven method to optimize thin-film processes in semiconductor manufacturing by integrating chemical vapor deposition with machine learning. By analyzing 91 process parameters and applying the analysis of variance and principal component analysis, we identified key variables influencing photoluminescence quality. A predictive model was built using the random forest (RF) algorithm and compared with the k-nearest neighbors method. The RF-based model demonstrated superior accuracy and robustness. The proposed method improves process stability, increases yield, and supports automated intelligent manufacturing across diverse material systems. Quantitatively, the RF-based configurations achieved R2 as high as 0.936 (mean intensity) and 0.842 (wavelength standard deviation (STD)), with mean squared error (MSE) minimized to 2.71 × 106 for mean-intensity prediction and 2.064 for wavelength-STD prediction, corresponding to up to a 96.7% MSE reduction relative to the k-nearest neighbors (KNN) baseline.