Sepsis is a noted cause of mortality in hospitalised patients. Early prediction of sepsis facilitates a better targeted therapy which in turn reduces patient mortality rate. This study has developed a methodology to allow automatic detection of sepsis 6 hours prior its clinical presentation. The initial solution developed in this study focuses on four vital signs namely: Heart rate (HR), temperature, Systolic Blood Pressure (SBP) and respiration rate. Additionally, we have considered the patient's age and the hours since ICU admission. Data from the available training set of 40,000 subjects were studied. Ninety percent of the data were used to train and validate the developed model. The remaining 10% were used to test the classifier. Missing measurement were replaced by the final measurement for each patient. Vital signs were automatically scored considering Modified Early Warning Scoring (MEWS) system. In addition, in order to address the changes in a patient's conditions after entering ICU, at each timestep, the difference between the measurement of the selected vital signs and the first hour in ICU were calculated and assigned a value -1, 0 and 1 based on reference to a median for each subject. The provided labels of sepsis were shifted 6 hours to use the measurements of each timestep to predict the sepsis label in 6-hour time. As the sepsis labels are hugely imbalanced, RUSBoost classifierer was used to first under-sample the dataset to produce a quasi-synthetically generated dataset and then classify it with decision trees. The proposed classifier has the accuracy of 0.95, normalised U-score of 0.48 with AUROC and AUPRC of 0.65 and 0.07, respectively. It is concluded that classifying the four scored vital signs obtained by using MEWS, with RUSBoost classifier can be a good method to predict sepsis for patience admitted in the ICU.