A Machine Learning Web Application for Real-Time Thermoelectric Property Predictions

Area of Science:

• Materials Science
• Computational Chemistry
• Data Science

Background:

• Machine learning (ML) is increasingly used in thermoelectric research for material discovery.
• Many existing ML studies lack practical application and are purely investigative.
• There is a need for accessible and cost-effective tools for predicting thermoelectric properties.

Purpose of the Study:

• To develop a resource- and cost-effective, single-page application machine learning framework for thermoelectric research.
• To provide a publicly accessible web interface for predicting thermoelectric properties.
• To offer an alternative rapid and affordable deployment option via a cloud-based function app.

Main Methods:

• Developed a single-page application (SPA) machine learning framework.
• Deployed the SPA using Docker containers in Microsoft Azure cloud.
• Hosted the static webpage on GitHub for public accessibility.
• Integrated ML models for predicting thermoelectric properties, including the power factor.
• Created an Azure Function App for an alternative deployment.

Main Results:

• A publicly accessible web application (https://kleinkeresearchgroup.github.io/TE_Predictions) was launched.
• The framework provides predictions for thermoelectric properties based on previous studies.
• An additional ML model was included for predicting the power factor of thermoelectric materials.
• An Azure Function App (https://fn-research-kleinke-app-jl-2025.azurewebsites.net/index) was developed as a deployment alternative.

Conclusions:

• The developed framework offers a practical and accessible tool for researchers in thermoelectric material discovery.
• The web-based interface and cloud deployment options enhance the utility and affordability of ML in this field.
• This work facilitates faster and more cost-effective prediction of crucial thermoelectric properties, advancing material science research.