Comparing the Efficacy of Machine Learning Models on Potentially Hazardous Objects

An asteroid or a comet is termed as a potentially hazardous object (PHO) if its orbit is such that it can closely approach the Earth and is huge enough to cause serious regional damage in the event of an impact. There are currently nearly 1,500 asteroids documented that have the potential to cause serious damage, but these are solely the ones we know about. In an event like this, it becomes a necessity to keep a check on all celestial bodies near earth so that we can take appropriate measures to minimize the damage. This project aims at analyzing data gathered by NASA about various asteroids or PHOs and predicting whether an Earth approaching body is hazardous or not. On finding an appropriate dataset, extensive data pre-processing was done on Machine Learning workspace on Azure. The Azure platform for machine learning algorithm provide a quick and easy way to apply machine learning algorithms as well as for modelling of data. Experimental results obtained by using diverse Machine Learning models were compared. The Two-Class Decision Forest provided the highest accuracy, a ROC-AUC score of 0.99.