Postgraduate Projects

Predicting the diameter of an asteroid is an ongoing challenge for identifying asteroids that could be potentially dangerous. Using a dataset from Kaggle (137,635 by 21 after data wrangling), a variety of models were implemented to further explore the diameter of the asteroids, including penalized regressions, support vector machines (SVM), naive bayes (NB), decision tree, random forests (RF), boosted trees (XGBoost) and neural network. A mixture of classification and regression tasks were implemented, with the regression models presenting lower error rates compared to the classification models. Overall, the RF and XGBoost were the most effective regression models for predicting the asteroid diameter, which improved further with hyperparameter tuning. Future research should prioritize ensemble methods that incorporate classification and regression approaches.