Predicting A Pulsar Star

Inspiration

I chose this dataset because I have a strong interest in astrophysics and doing a project like this is similar to some astrophysics research. In fact, the research I already do is strongly concerned with radio frequency interference (RFI), so it would be interesting to see how RFI is dealt with in different experiments within the same field.

What it does

Based on eight parameters, we want to predict whether a signal is RFI or a real pulsar star. Essentially, we have nine coloumns and we want to be able to predict the entries of the ninth – is it a 0 (RFI) or 1 (real star)?

How I built it

Starting with three models, a Support Vector Machine, k-Nearest Neighbours Classifier and a Random Forest Classifier, I ran tests with a whole range of different hyperparameters, and combinations of these. Based on performance on training set and validation set, the Random Forest Classifier was chosen.

Results and in-depth discussion can be found in the poster and the github repository!

Built With

• jupyter-notebook
• python