<p>This study proposes a comprehensive analytical strategy that integrates three-dimensional excitation-emission matrix (3D-EEM) spectroscopy, parallel factor analysis (PARAFAC), and machine learning algorithms, aiming to achieve accurate prediction of the contents of dual components in Perilla frutescens. The methodology encompasses several key steps: (1) Spectral preprocessing, including the removal of Rayleigh and Raman scattering via Delaunay triangular interpolation, followed by signal enhancement using Savitzky-Golay (SG) convolutional smoothing; (2) Spectral feature extraction through PARAFAC, with component identification validated by residual sum-of-squares analysis and split-half validation, successfully characterizing two dominant fluorescent constituents—chlorophyll and flavonoids; (3) Development of quantitative prediction models based on Random Forest (RF) and Support Vector Machine (SVM) algorithms. The experimental results indicated that the RF model exhibited excellent predictive performance and generalization ability for both chlorophyll and flavonoid contents. Specifically, for the chlorophyll prediction model, the coefficient of determination (R²) values of the training set and test set were 0.973 and 0.967, with the Root Mean Square Error (RMSE) values being 0.051 and 0.006, respectively. For the flavonoid prediction model, the R² values of the training set and test set were 0.966 and 0.944, while the RMSE values were 0.051 and 0.007, respectively. These findings demonstrate the efficacy of the proposed 3DEEM-PARAFAC-RF framework for analysis of plant bioactive compounds. This methodology not only offers a robust technical approach for quality evaluation of Perilla frutescens, but also provides a valuable reference for quantitative analysis of phytochemicals in complex botanical matrices.</p>