machine-learning-challenge

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, you will create machine learning models capable of classifying candidate exoplanets from the raw dataset. In this homework assignment, you will need to:

• Preprocess the raw data
• Tune the models
• Compare two or more models

dataset

https://www.kaggle.com/nasa/kepler-exoplanet-search-results

column definition

https://exoplanetarchive.ipac.caltech.edu/docs/API_kepcandidate_columns.html

Results

• SVM model (svm_model.ipynb)

  • score: 0.74 (linear kernel)
  • confirmed cases only
    • kernel, precision, recall
    • linear, 0.73, 0.90
    • rbf, 0.53, 0.95 (can imporve to 0.60 with GridSearch)
    • poly, 0.51, 0.95

• Deep Learning model (deep_model.ipynb)

  • confirmed cases: loss = 70%; accuracy = 68%

• Logistic Regression (logistc_reg_model.ipynb)

  • score: 77.13%

• Decision Tree model (tree_model.ipynb)

  • score: 68%

• Random Forest model (random_forest_model.ipynb)

  • score: 77.34% (with GridSearch improvement)
  • feature importance
    • [(0.1686855604159887, 'koi_model_snr'),
    • (0.16162502363875983, 'koi_prad'),
    • (0.10905044931577702, 'koi_period'),
    • (0.10568093894563921, 'koi_impact'),
    • (0.10425958378206863, 'koi_depth'),
    • (0.10027461918949472, 'koi_duration'),
    • (0.09300511042263368, 'koi_insol'),
    • (0.07963441337955753, 'koi_teq'),
    • (0.07778430091008075, 'koi_time0bk')]

Conclusion

• Random Forest has the best score of 77.34%. It is the best model based on my testing.
• Random Forest is useful to point out the top 2 most important features are "koi_model_snr" and "Koi_prad".