Contributors: Krieger Benjamin, Zell Kilian, Jegou Sam

Competition: Higgs Boson Machine Learning Challenge

Our best model was a ridge regression, with a polynomial augmentation of degree 5 and some other feature augmentations.

All the outliers were set to the mean of their corresponding features, before the standardization process (0 mean, 1 standard deviation).

It has been tried to remove the features that had too much outliers (~90%), or separate the dataset in function of the indicator variable JET, but it decreased our accuracy.

Our model is a simple ridge regression, with regularization of $\lambda = 10^{-5}$. The input data has been augmented via:

• a polynomial augmentation of degree 5
• some other augmentations for the first features, with cosines, sinus, sinc, and others to add non linearity

Then we project the prediction on {-1,1} with a threshold at $0.05$.

Put the files run.ipynb and implementations.py in the main folder.

In the same folder, create a folder 'Data' containing the data file 'sample-submission.csv'.

Run the file run.py. It will generate in your current folder a file 'submission.csv', containing the prediction of our model.

Machine learning course at EPFL:

Description of the dataset: https://higgsml.lal.in2p3.fr/files/2014/04/documentation_v1.8.pdf