Background: Sepsis is a potentially fatal condition that results from the body’s dysregulated response to infection. If left untreated, sepsis causes tissue damage and organ failure which may ultimately result in death. To improve patient outcomes, sepsis needs to be detected earlier, thereby prompting rapid initiation of the sepsis bundle. Methods: The PhysioNet 2019 challenge aggregated electronic medical record data from two clinical sites, consisting of data from over 40,000 admissions. The challenge implemented the Sepsis-3 definition, which included the sequential organ failure score as a major component. Using an approximated timestamps of sepsis onset, defined by the challenge organizers as ‘tSepsis’, we implemented a random forest model to estimate the probability of a patient developing sepsis. Using fixed-width overlapping time-windows of 6 hours, we extracted a number of temporal and statistical features from the challenge dataset. Our random forest model used 700 trees and the Gini impurity index for information gain. Results: We established a probability criterion of 0.3 to generate alerts, in order to maximize a normalized score established by the PhysioNet Challenge team. Using the aforementioned method, a 5-fold cross validation utility of 0.812 was obtained. Top three predictive features (in order) were the variance in heart rate, maximum temperature, and skewness in heart rate. Conclusion: Predictive models can augment clinical intuition of the bedside staff to improve sepsis detection. In this paper, we implement an interpretable machine learning model to support earlier detection of sepsis. This can potentially guide clinicians on an earlier initiation of the sepsis bundle in order to reduce morbidity and mortality.