Machine Learning - Exoplanet Exploration

Background

Over a period of nine years in deep space, the NASA Kepler space telescope has been out on a planet-hunting mission to discover hidden planets outside of our solar system.

To help process this data, I created machine learning models capable of classifying candidate exoplanets from the raw dataset.

Steps

Preprocessed the Data

• Preprocessed the dataset prior to fitting model.
• Performed feature selection and removed unnecessary features.
• Used MinMaxScaler to scale the numerical data.
• Separated the data into training and testing data.

Tuned Model Parameters

• Used GridSearch to tune model parameters.
• Tuned and compared at least two different classifiers.

Findings

The following machine learning models were used to process the dataset:

• Decision Tree
• K-Nearest Neighbors
• Logistic Regression
• Neural Network
• Random Forest
• Support Vector Machine

After feature selection and tuning the parameters of each model, the Random Forest model classifies candidate exoplanets most accurately. The Random Forest model data score was higher before tuning than the other models after tuning.

With additional feature selection and parameter tuning, the Random Forest model would be good enough to predict new exoplanets.

Random Forest

Testing Data Score (after feature selection): 0.9010297482837528

Neural Network

Testing Data Score (after feature selection): 0.885011434555

Support Vector Machine

Testing Data Score (after feature selection and parmeter tuning): 0.8775743707093822

Logistic Regression

Testing Data Score (after feature selection and parmeter tuning): 0.8747139588100686

Decision Tree

Testing Data Score (after feature selection and parmeter tuning): 0.8735697940503433

K-Nearest Neighbors

Testing Data Score (after feature selection and parmeter tuning): 0.8050734312416555