machine-learning-challenge

Background

The purpose of this project is to discover hidden planets using NASA keplar space telescope data and machine learning in Python. In exoplanet_data.csv, there are various columns of telescope data and more detailed description can be found in the NASA Exoplanet Archieve Website.

The project was done in Python, but there are multiple libraries like that sklearn, tensorflow, pandas and joblib were used.

Analysis

Two models were used to compare the results of accuracy of the data.

1. Sequential Model

1-1. First model - sequential model with two layers

The sequential model was built using tensorflow library and the accruacy was 0.890 with a loss of 0.265.

1-2. Prediction and actual data

Using the sequential model, koi disposition was predicted as shown below. The predicted data matched with the actual data.

2. Linear Model

2-1. Second model - linear model

The linear model was built using sklearn library and the accracy of testing data was 0.856 and the training data was 0.837

2-2. Hyperparameter tuning

The best parameter and accuracy was found using GridSearch model in sklearn library and the best parameter and the score were 0.0001 and 0.878

2-3. Classification report

The classification report of the linear model is shown below: