In recent years, the diverse aglycones of Scutellaria baicalensis Georgi (SBG) have attracted growing interest owing to their broad pharmacological activities. As a crucial step in the research of these flavonoids, extraction critically influences their applicability in drug development and therapeutic performance. Efficient extraction strategies not only enhance yield and purity but also better preserve bioactivity, thereby helping to achieve expected effects in anti-inflammatory, antioxidant, and antitumor applications. However, current challenges include relatively low extraction efficiency of aglycones and unclear mechanisms against oral squamous cell carcinoma (OSCC). Therefore, this study aimed to optimize the extraction process of aglycones, establish content prediction models, and preliminarily investigate the anti-OSCC mechanisms. For extraction optimization, endogenous enzymatic hydrolysis (EEH) combined with ultrasound-assisted extraction (UAE) was applied, and parameters were systematically optimized using an integrated AHP-CRITIC weighting method coupled with Box-Behnken design-response surface methodology (BBD-RSM). For content prediction, infrared spectroscopy combined with chemometrics was employed to develop partial least squares regression (PLSR) models for four key aglycones: baicalein, wogonin, norwogonin, and oroxylin A. For pharmacological investigation, UPLC-Q-TOF/MS, network pharmacology, molecular docking, and experimental validation were integrated to explore the anti-cancer effects. The results demonstrated that under optimal extraction parameters-enzymatic hydrolysis at 55 °C for 10 h, liquid-solid (L/S) ratio of 40:1, ultrasonic time of 20 min, pH 5.4, and ultrasonic power of 350 W-the extraction efficiency was significantly improved. The established prediction models performed reliably, with Rpre > 0.77, RPD > 2.1, and a mean RSEP of 8.71 across all four aglycones. Moreover, baicalein, 4'-hydroxywogonin, norwogonin, skullcapflavone II, and luteolin were identified as core active components that induce apoptosis in CAL-27 cells by regulating targets including EGFR, BCL-2, and BAX, and modulating the PI3K-AKT signaling pathway, thereby exerting anti-OSCC effects. In conclusion, the optimized extraction process is efficient and reproducible, the content prediction models show good predictive ability, and the anti-OSCC mechanisms are supported by experimental evidence, providing valuable insights for the development of natural medicines.