Predicting atomic-level reaction mechanisms for S_N2 reactions via machine learning

Area of Science:

• Chemistry
• Materials Science
• Computational Science

Background:

• Understanding atomic-level reaction mechanisms is crucial for advancing chemical sciences.
• Predicting dynamical processes in structural evolution requires sophisticated computational approaches.

Purpose of the Study:

• To develop and validate a machine learning model for predicting atomic-level reaction mechanisms.
• To assess the impact of data class granularity on model performance and convergence.

Main Methods:

• A joint-voting model integrating three parallel machine learning algorithms was developed.
• The model was trained on 1700 trajectories, with experiments using ten, seven, and five classes.
• Model convergence and prediction accuracy were evaluated across different class divisions.

Main Results:

• The five-class model demonstrated the fastest convergence and highest prediction accuracy (92.0%).
• Prediction success rates increased as the number of trajectories per class increased.
• Model convergence was achieved with 100, 100, and 70 trajectories for the ten-, seven-, and five-class experiments, respectively.

Conclusions:

• Machine learning effectively predicts elementary dynamical processes in structural evolution.
• The proposed joint-voting model advances the prediction of atomic-level reaction mechanisms.
• Optimizing data stratification enhances the efficiency and accuracy of machine learning models in chemistry.