Objective <p>To address limitations in pulse oximetry accuracy associated with low saturation and melanin content using machine learning (ML) and to compare our results to standard pulse oximetry and gold standard arterial blood gas readings.</p> Background <p>Oxygen saturation is traditionally measured through the gold standard arterial blood gas (SaO₂) or pulse oximetry (SpO₂), which approximates SaO₂ using light absorption patterns. However, SpO₂ has been shown to overestimate oxygen saturation, particularly in individuals with darker skin tone, leading to hidden hypoxemia and delayed medical interventions.</p> Method <p>We developed a machine learning (ML) model trained on the BOLD dataset, integrating patient data from eICU, MIMIC-III, and MIMIC-IV (<i>n</i> = 49,093). With 64 clinical features and 2 outcomes (SaO₂ and hidden hypoxemia events), we trained regression ML models (linear regression, random forest, and XGBoost) to predict SaO₂, minimizing the mean squared error between predicted SaO₂ and ground truth SaO₂. We used our test data to compare model performance to standard SaO₂ measurement with accuracy root mean square error (A_rms), R², and change in hidden hypoxemia events. We used SHapley Additive exPlanations (SHAP) to rank important features for SaO₂ prediction.</p> Results <p>The XGBoost-Vanilla model improved A_rms to 3.3% from a baseline of 4.1%. In low SaO₂ patients (SaO₂ &lt; 90%), the accuracy of pulse oximetry was heavily compromised with A_rms of 10.3%; the linear regression model with weighted loss was able to reduce A_rms to 8.6%. We found that SpO₂, creatinine level, mean corpuscular hemoglobin level, respiratory rate, and age at admission were the leading features driving the SaO₂ prediction.</p> Conclusion <p>These findings suggest that ML-based models can enhance the accuracy of the standard SpO₂. Further investigation is warranted to address SpO₂ inaccuracies and bias.</p>