Small-Scale Machine Learning Comparison Applied to Predict Bandgap and Photocatalytic Performance of Rare-Earth-Doped Spinel Ferrites

1/ AI 활용 방법: Comparison of regression models for small data sets (ridge regression, random forest, extra trees, gradient boosting)
2/ 적용 과제: Prediction of the optical bandgap and photocatalytic performance of rare-earth-substituted spinel ferrites
3/ 보고된 결과: random forest recorded the lowest error among the candidate models, but MAE 21.43 percentage points and R² = -0.56 indicated limited generalization performance
4/ Scientific Reports compared regression models for small data sets to predict the optical bandgap and photocatalytic performance in rare-earth-doped spinel ferrites.
5/ The researchers examined ridge regression, random forest, extra trees, and gradient boosting, and reported that random forest recorded the lowest error among the candidates.
6/ The materials-science significance of this issue is that it is an attempt to estimate optical and photocatalytic properties from relatively simple tabular features in composition-modified oxide ferrite systems.
7/ However, the reported performance of MAE 21.43 percentage points and R² = -0.56 suggests that the predictions did not sufficiently generalize experimental trends.
8/ In practice, one can collect material composition, doping level, processing conditions, and measured bandgap and photocatalytic indices, then compare multiple regression models using the same split and the same preprocessing.
9/ However, because this result comes from a comparison on limited data, it does not mean it can be directly extended to other composition spaces or experimental conditions. [1]