Predicting academic-challenge success

Abstract

Academic competitions and challenges comprise an effective mechanism for rapidly advancing the state of the art in diverse research fields and for solving practical problems arising in industry. In fact, academic competitions are increasingly becoming an essential component of academic events, like conferences. With the proliferation of challenges, it is becoming more and more relevant to distinguish potentially successful challenges before they are launched. This in order to better allocate resources, time slots, sponsorship and even to have a better estimate of expected participation. This paper presents a first study in this direction: We collected a data set from Kaggle and aim to predict challenge success by using information that is available before a competition starts. We characterize competition proposals by textual information and meta-features derived from information provided by organizers, and use these features to predict challenge success (estimated by the number of participants and submissions). We show that both, text and meta-features convey predictive information that can be used to estimate the success of an academic challenge.