Detecting exoplanets using convolutional neural networks and support vector machine

This repository is connected to a bachelor thesis with the purpose of finding exoplanets with two machine learning algorithms. The bachelor thesis can be found here. For this study two machine learning algorithms, support vector machine and convolutional neural networks, where developed and evaluated on a data set to determine which algorithm that performed the best. This project was implemented in pyhton, executing it in Google Colab on a GPU.

Data set

The data set can be found at a host at kaggle.com. It contains time series of light intensity of a star which binary classified in two catagories, "star with no exoplanet in orbit" or "star with an exoplanet in obrit". This data-folder on Google Drive contains all the data sets used for training and testing the SVM model.

Code

CNN.py

• Containing code for:
  • Preprocessing of data
  • Training of the network
  • Validation of the network

CNN_loadmodel.py

• Code for using an already trained model, saved on your computer

graphics_preprocessing.py

• Creates following graphics:
  • Preprocessing steps for CNN
  • Raw data of light curves for the thesis

SVM.py

• Containing code for:
  • Preprocessing of data
  • Training of the network
  • Validation of the network

SVM_gridsearch.ipynb

• Containing code for:
  • Varying the model-parameters with a feature set as input data
  • Evaluating the parameters by using cross-validation

SVM_nofeat_gridsearch.ipynb

• Containing code for:
  • Varying the model-parameters with the raw data set as input
  • Evaluating the parameters by using cross-validation

SVM_preprocess.py

• Code for different preprocessing options for the SVM method

Required packages

• Keras
• Pandas
• Numpy
• Matplotlib

Code authors

Sofia Dreborg: @sofiadreborg

Maja Linderholm: @majalinderholm

Jacob Tiensuu: @JacobTiensuu

Fredrik Örn: @fredrikorn