Farthest point sampling in property designated chemical feature space as an effective strategy for enhancing the machine learning model performance for small scale chemical dataset

Machine learning (ML) model development in chemistry and materials science often grapples with the challenge of small and imbalanced labeled datasets, a common limitation in experimental studies. These dataset imbalances can precipitate overfitting and diminish model generalization. Our study explores the efficacy of the farthest point sampling (FPS) strategy within targeted chemical feature spaces, demonstrating its capacity to generate well-distributed training sets and consequently enhance model performance. We rigorously evaluate this strategy across various ML models, including artificial neural networks, support vector machines, and random forests, using datasets with target physicochemical properties such as standard boiling points and enthalpy of vaporization. Our findings reveal that FPS-based models consistently surpass randomly sampled models, exhibiting superior predictive accuracy and robustness, alongside a marked reduction in overfitting. This improvement is particularly pronounced in smaller training set, attributable to increased diversity within the training data’s chemical feature space. Consequently, FPS emerges as an effective and adaptable approach for achieving high-performance ML models at reduced cost by limited and biased experimental datasets typical in chemistry and materials science.