Predicting Carbon Dot Photoluminescence: A Comparative Machine Learning Study on Systematic Synthesis Data.

Area of Science:

• Materials Science
• Nanotechnology
• Computational Chemistry

Background:

• Designing carbon dots (CDs) with specific optical properties is challenging due to complex synthesis and unpredictable relationships.
• Current methods often involve lengthy trial-and-error processes, hindering rapid material development.

Purpose of the Study:

• To develop a data-driven approach using machine learning to predict the photoluminescent emission of carbon dots (CDs).
• To accelerate the design and synthesis of CDs with tailored optical properties by predicting outcomes from synthesis parameters.

Main Methods:

• Collected a dataset of 407 carbon dot syntheses using p-benzoquinone and ethylenediamine in various solvents.
• Applied and compared ensemble learning algorithms: Random Forest, XGBoost, and CatBoost.
• Evaluated model performance using the coefficient of determination (R²).

Main Results:

• CatBoost demonstrated superior predictive accuracy for carbon dot photoluminescence.
• Achieved a mean cross-validation R² of approximately 0.98, outperforming Random Forest and XGBoost.
• Validated the efficacy of gradient boosting algorithms for modeling chemical synthesis data.

Conclusions:

• Machine learning, particularly CatBoost, offers an efficient computational tool for predicting carbon dot optical properties.
• This data-driven approach can guide the on-demand synthesis of functional nanomaterials.
• Highlights the potential of AI in accelerating materials discovery and design.