Data-Driven Strategies for Winning IoT Challenges

Case Study

Abstract

In the competitive world of Internet of Things (IoT) endeavors, predicting a project’s success in contests can be challenging due to subjective and varied judging criteria. Our study addresses this problem by using machine learning to analyze outcomes in IoT contests, focusing on 104 competitions with a total of 5,863 projects from the Hackster.io platform. To the best of our knowledge, this is the first study to address this problem in the IoT community. We evaluated seven different machine learning models, which revealed that ensemble methods, such as random forest and gradient boosting, were the most effective, achieving an average accuracy of 80% and an Area Under the Curve (AUC) of 77%. We also performed a mixed-effects logistic regression that not only predicts a project’s likelihood of winning with an AUC of 86% but also uncovers the most significant factors that increase a project’s chances of success. These insights are valuable for IoT project creators, providing them with practical advice on how to improve their projects. Our research also offers useful insights for other stakeholders in the IoT community, such as IoT engineers and contest organizers, helping them understand what makes projects more likely to win. This study is a significant step in helping participants in IoT contests make informed decisions and increase their chances of success.