Rationale: Acute symptomatic seizures (ASyS) are a contributor to morbidity in hospitalized patients. Early recognition of patients who may be at risk for developing seizures would be useful. Methods: Single-center retrospective cohort analysis of EEG data in adult patients who underwent cEEG monitoring at a large tertiary referral institution between 2010 and 2019. Patients who were seizing at the onset of recording were excluded. A total of 43,542 patients met the inclusion and exclusion criteria. A subgroup of patients limited to seizures within 72 hours, or the first 72 hours of cEEG monitoring if there were no seizures, were used for static model development. EEG data was classified based on presence or absence of the following features: CS (continuous slowing), NL (normal), GRDA (generalized rhythmic delta activity), BS (background suppression), IS (intermittent slow; generalized, lateralized or regional), CRS (generalized continuous rhythmic slow), SW (sharp wave), SP (spike), SWC (spike wave complex), SSWC (slow spike wave complex), PSP (polyspikes), LPD (lateralized periodic discharge (previously known as periodic lateralized epileptiform discharges (PLEDs)), BIPDs (bilateral independent periodic discharges (BIPDs), Generalized Periodic Discharges (GPDs) (including GPD-unclassified, GPD-TM (triphasic morphology), GPD-SC (sharply contoured), and GPD-R (rhythmic)), BSUP (burst suppression), SBG (slow background), ECI (electrocerebral inactivity), pattern coma (PCO) (including spindle coma, delta coma, alpha coma, theta coma, and beta coma), EXF (excessive fast), ASY+ (asymmetry increased), and ASY- (asymmetry decreased). Machine learning models based on day-to-day dynamic EEG changes (LSTM) and snapshot static EEG features (XGBoost) over the prior 72 hours or until seizure were applied to evaluate their ability to predict seizure occurrence. Results: One thousand nine hundred sixty-eight (4.5%) patients eventually experienced a seizure during their EEG recording. In the subgroup of patients who were analyzed using static model, patients had an average age of 60.9 ± 17.3 years (mean ± SD) were monitored for 37.6±20.1 hours (mean ± SD). In this cohort, 1866 (5.1%) of these patients experienced seizures. The static model was able to predict seizure occurrence based on cEEG data with sensitivity and specificity of 0.657 and 0.799 respectively with an area under the curve (AUC) of 0.728. The most important EEG features used in this model were analyzed using shapley additive explanations (SHAP) analysis. These are summarized in Figure 1 and some of the most important features increasing seizure risk included presence of LPD, PSP, BSUP, SBG, SW, SP, GPD-Unclassified, BIPD, SWC, EXF, GPD-SC, and PCO. The dynamic model was able to predict seizure occurrence with sensitivity and specificity of 0.825 and 0.662 respectively with an AUC of 0.805. Conclusions: Machine learning models could be applied to cEEG data to predict seizure occurrence. Dynamic day-to-day EEG data are more useful in predicting seizure than snapshot static EEG data. These models could potentially be used by clinicians to make informed decisions regarding duration of cEEG monitoring. Funding: Please list any funding that was received in support of this abstract.: N/A