Description

Comparison of Models using NASA Kepler data

Target

'koi_disposition' - The disposition in the literature towards this exoplanet candidate. One of CANDIDATE, FALSE POSITIVE, NOT DISPOSITIONED or CONFIRMED (i.e., likelihood that a given exoplanet is a true planet)

Resources

• Exoplanet Data Source
• Scikit-Learn Tutorial Part 1
• Scikit-Learn Tutorial Part 2
• Grid Search

Preprocessing

• Preprocess the dataset prior to fitting the model.
• Perform feature selection and remove unnecessary features.
• Use MinMaxScaler to scale the numerical data.
• Separates the data into training and testing data with TrainTestSplit.

Currently Supported Models

• Decision Tree - uses GridSearch to tune model parameters. [K-Nearest Neighbors, Support Vector Machine, Recurrent Neural Network - Keras/Tensorflow]

Denouement

1 . Decision tree ensemble (random forest) points to solar mass centroid offset of exoplanet mass as strongest predictor of planetary viability.