With the upcoming space missions and surveys, detailed analysis of exoplanets and their host stars can be very time-consuming. This motivated us to use machine learning to automate this process in an accurate and precise manner.
What it does
Our AI algorithm predicts the physical properties of an exoplanet, such as its rotational period, using its transit light-curves.
How we built it
We used Python in all parts of our codings. At first, we generated several synthetic light-curves and created a large training dataset. Then we created a model using our training set via the Convolutional Neural Network (CNN) technique. Finally, we checked the performance of our algorithm using both the synthetic data and the real data from the Kepler mission in a test set.
Challenges we ran into
The biggest challenge for us was the preprocessing step of our data. We had to create a homogeneous synthetic dataset and then we had to make the real data compatible with that.
Accomplishments that we're proud of
We used machine learning techniques to estimate multiple physical features of extra-solar systems that host planets. The comparison between our results and the reference results was extremely consistent. In addition, the determination process is approximately %98 faster than the other known techniques that make this work even more unique.
What we learned
We learned about many different aspects of machine learning and coding in this project. In particular the challenges we had, let us find the potential challenges in such works and also helped us learn how to deal with them.
What's next for AI Exo-plorer
We have many plans!! First of all, we believe we can increase the number of features our AI Exo-plorer predicts. For example, we can add the other physical properties of the host star and see if our algorithm can find a correlation between the data and these newly added features.
We also are planning to apply the AI Exo-plorer on a wider range of real light-curves from the TESS mission.