Prediction of Mefenamic Acid Crystal Shape by Random Forest Classification

Area of Science:

• Materials Science
• Crystallization Engineering
• Computational Chemistry

Background:

• Particle shape significantly influences bulk material properties.
• Controlling crystal morphology is crucial for material performance.
• Predicting crystal shape from crystallization conditions remains a challenge.

Purpose of the Study:

• Develop and apply machine learning models to predict mefenamic acid crystal shape.
• Identify key factors influencing crystal morphology during recrystallization.
• Enhance the predictability of crystal habit formation.

Main Methods:

• Recrystallization of mefenamic acid in 30 organic solvents.
• Dataset generation including solvent descriptors, process conditions, and crystal shape.
• Training and validation of Random Forest classification models.

Main Results:

• Achieved up to 93.5% prediction accuracy for crystal shape.
• Models using only solvent physical properties and supersaturation showed higher accuracy.
• Discovery of a new mefenamic acid solvate in poorly predicted cases.

Conclusions:

• Random Forest models reliably predict mefenamic acid crystal morphology in most solvents.
• Solvent physical properties are key predictors of crystal habit.
• Additional factors are needed for a generalized predictive morphology model.