<p>To enable real-time prediction of rock mechanical parameters during drilling, this study proposes a method based on vibration-while-drilling (VWD) spectral features. Three-component vibration signals were collected experimentally, and their dominant frequency, low-frequency energy, and spectral centroid were extracted as predictors. A ridge regression model was developed to map these spectral features to rock mechanical parameters. Compared with conventional drilling parameters, the spectral descriptors capture lithology-dependent stiffness more effectively, with the low-frequency (0–20 Hz) energy showing a strong correlation with rock strength. Validation on tuff specimens achieved high accuracy (mean R² &gt; 0.80) and stable calibration between predicted and measured values. Bootstrap and permutation analyses confirmed the consistency and interpretability of feature contributions, while ridge-penalty scanning demonstrated strong resistance to overfitting. The proposed approach provides an efficient and interpretable framework for real-time field prediction of rock mechanical parameters and offers a foundation for multi-lithology and physics-informed model extensions.</p>