Predict Transparent Semiconductors using Machine Learning

Transparent semiconductors may be the future for flat panel display!

So what determine the transparency of the semiconductors?
Short answer: The band gap.
Then what determine the stability of the transparent semiconductors?
Short answer: The formation energy.

This kaggle competition predicts the band gap and the formation energy of 600 semiconductors, given structural properties of 2400 semiconductors as the training set.

Key Results/Highlights:

1. Predicted bandgap energy is used to predict the formation energy, as they are highly correlated.
2. CV score on formation energy prediction is increased by a few percentages when switched over to gradient boosting regression, due to the high bias (train score > CV score) training by random forest regression.
3. Percentage of In and Al are important to determine the bandgap, while percentage of Ga is important to the formation energy.

Results:
Accuracy = 95.0 %

• Click here for exploratory data analysis.

• Click here for features selection based on correlations.

• Click here for machine learning application.

Visit my github page for more information.