Material Prediction and Design

• 

  Some results

• 

  Some results

• 

  Properties correlation

• 

  Elements correlation

Inspiration

Data science and Machine Learning is an up-and-coming field and the talk of the town when it comes to the tech world. In the past few years, ML-based approaches to real-world problems have become popular. Predicting the composition of a material based on its properties is one such problem. This project presented me with an opportunity to work at the intersection of the two domains dearest to me. AI and Physics.

What it does

• Predicts the properties of a material based on its composition.
• Predicts the composition of a material based on its properties.

How I built it

We were provided with a sample code for the task of predicting the properties of a material based on its composition. Our task was to improve the output of this code as well as to create a model to predict the composition of a material based on its properties. It did this by trying various Machine Learning models and picking the one with the least error on training and testing data and the best R^2 score. This was done using Python Libraries PyTorch and Sklearn.

Challenges faced

The first and obvious challenge was the size of the data set. A small training set restricts the model from learning well. The second challenge was the missing data in the data set. This was handled by replacing the missing data with the appropriate values as mentioned in the python notebook.

Accomplishments

I was successfully able to predict the composition of materials based on their properties. This project allowed me to showcase my current range of knowledge in Machine Learning and apply it to a real-world problem.

Built With

• numpy
• pandas
• python
• pytorch
• sklearn