A Comparative Study of Different Boosting Algorithms for Predicting Olympic Medal

Abstract

Predicting whether an athlete is likely to win a medal in the Olympic games is new. The studies on Olympic Games are mostly trying to predict the total medals of a nation possible to achieve or a country’s performance by applying statistics approaches. Some works even expand the data utilized for medal predicting by including more years and predictor factors such as country host as well as increasing the level of data granularity. Machine learning, in particular boosting algorithms, has had a massive influence in improving the accuracy of prediction models. To accurately classify an athlete, three different machine learning approaches can be utilized. In this study, three separate boosting algorithms, namely Light Gradient Boosting Machine (LightGBM), extreme Gradient Boosting (XGBoost), and Category Boosting (CatBoost) are evaluated using Olympic historic dataset, first with default parameters, then with hyperparameters by applying Grid Search algorithm. Four different types of performance evaluation metrics were computed with 5-fold Cross-Validation (CV) approach. The best results were obtained with the XGBoost approach on hyperparameters, achieving an accuracy of above 90%, a precision of 96.8%, and a recall of 83.2%.