Aims: Early prediction of sepsis is very crucial in detecting and safe guarding the patients against the disease. While late prediction of sepsis is life threatening, predicting sepsis in non-sepsis patients results in shortage of hospital resources. This research aims at predicting the disease very early with the help of artificial neural networks. Methods: The strategy is to apply Gradient Boosting Classifier (GBC) on the data and find out the important features that make a decision in predicting sepsis. Then apply only these features to a deep neural network to improve accuracy and to improve the scalability, given huge amount of data. Around 761995 hours of physiological data is taken for making predictions. Each row which contains at-least 10 non-null features from 40 features was selected. Among those the remaining null values are filled with the mean value of that particular column values. The strategy applied for filling up the null values worked very well and gave rise to high accuracy scores. Then important features are derived from the data with the help of classifier’s feature importance tool. The next step is to feed only relevant features as inputs to the deep neural network model for scalability and to achieve more accuracy. Results: GBC machine learning model is able to achieve a high accuracy score of 98.6% and 98.5% on the training and test data sets respectively. When given data of single subject, the algorithm predicted the results correctly with high probabilities. Conclusion: Since Gradient Boosting Classifier model performed well for this physiological data, this model can be recommended to make an early prediction of the disease thereby reducing the mortality rate.