The next big thing in science

This paper presents an approach to predicting the future development of scientific research based on scientific publications from the past two centuries. We have applied machine learning methods on the Microsoft Academic Graph dataset of scientific publications. Our experimental results show that the best performance is obtained for a noticeable increase of the topic frequency in the last 5 years compared to the previous 10 years. In this case, our model achieves precision of 74.3, recall of 71.7 and F1 of 73.0. Some topics that our model identified as promising are: proton proton collisions, higgs boson, quark, hadron, mobile augmented reality, variable quantum, molecular dynamics simulations, hadronic final states, search for dark matter.