Asteroid hunter received a Global Nomination.
For this challenge, we invite you to become "virtual contributors" to the Asteroid Grand Challenge and develop a hypothetical method, concept note or simple prototype that demonstrates how Machine Learning could be used to help us avoid the same fate as the dinosaurs.
Firstly, we looked for the dataset over several NASA website and also NASA SPACE APPS's repository . These data weren't homogeneous so that we had to reorganize and recode them.
Then, we investigated about the variables and their astrophysical meaning. This was a hard work due to the fact that none of the members of the group were astrophysicist.
After that, we selected the most useful variables for our predicting model and machine learning algorithms.
According to the models, we tested a large variety of them:
- Deep Neural Networks based on ELM, MLP-ELM, Random Forest
- SVM, MLP
The best results were obtained using MLP-ELM.
On the other hand, for expressing these results in an intelligible form, we have visualized them by using the python's package "poliastro". This package allows us to plot not only the orbit around the Earth of the asteroid but also its position relating to the Earth and other asteroids.
All this procedure (MLP-ELM + visualization) is able to classify new daily asteroid data into two possible categories: PHA (potentially hazardous asteroid) or not PHA.
We have also developed a Twitter bot which will daily inform about the NEO'S and PHA's objects.
Python and R programming languages, Jupyter, Tensorflow and Twitter