Exoplanets and its habitability, by Anna Vidal Perez

Overview

Exploratory analysis of Kepler's telescope detections and habitability of confirmed exoplanets. Classification of Kepler's detections into exoplanets or false positives with machine learning.

Ideas and hypothesis

Having a look at the data we could ask the following questions:

• How many of Kepler's telescope detections end up being confirmed exoplanets?
• What is the ratio of habitable planets of the total of confirmed exoplanets?
• What are the main characteristics of the confirmed exoplanets and their stars?
• Can we prove Kepler's third law of planetary motion (Law of Periods)?
• Can we use machine learning to classify if a detection is an exploanet or a false positive? And to classify the type of exoplanet?

Datasets

The first dataset is the Kepler Exoplanet Search Results from Kaggle and has information about the objects detected by the Kepler Space Observatory. This dataset will be used for machine learning classificiation.

• Shape: (9564, 50)
• Size: 3.7 MB
• Complexity: https://exoplanetarchive.ipac.caltech.edu/docs/API_kepcandidate_columns.html

The second dataset is also public and downloaded from Planetary Habitability Laboratory (PHL), University of Puerto Rico. It has more than 3000 already confirmed planets and contains interesting information about their possible habitability. It is the dataset used for the exploratory analysis of planet habitability.

• Shape: (3873, 68)
• Size: 1.6 MB
• Complexity: http://phl.upr.edu/projects/habitable-exoplanets-catalog/data/database

Data Analysis

1. Hypothesis testing to prove Kepler's third law of planetary motion. The law states:

"The square of the orbital period of a planet is directly proportional to the cube of the semi-major axis of its orbit."

Null hypothesis: the law of periods is not met.  
Alternative hypothesis: the law of periods is met.

Calculated correlation, Pearson's R and R squared.

2. T-test for significance between different types of planets.

Machine Learning

Random forest, KNN and logistic regression to classify Kepler objects into exoplanets or false negatives.

Accuracy scores:

• KNN: 0.81
• Logistic regression: 0.80
• Random Forest: 0.89
• Random Forest with number of parameters = 100: 0.90

Conclusions

• Kepler's third law of periods is valid.

• Habitable planets are very rare.

• We can classify a Kepler's telescope detection into exoplanet or false positive 0.81% of the time.

• Not data enough to classify the habitability of the exoplanet.

Slides

https://docs.google.com/presentation/d/12wC4PkwdBHOSofVmU3FqU4M3JNu8dGowPsvZTVexUAQ/edit?usp=sharing