Advanced predictive modelling of electric quadrupole transitions in even-even nuclei using various machine learning approaches

Empirical predictions of electric quadrupole transition probabilities, B (E2; 0*-> 2*), in even-even nuclei, are among the principles needed to solve the nuclear structure and collective behaviour. In this study, nine different ML algorithms, gradient boosting machine (GBM), random forest (RF), convolutional neural network (CNN), k-nearest neighbour (KNN), CatBoost, extreme gradient boosting (XGBoost), neural network (NN), support vector machine (SVM) and multiple linear regression (MLR), are evaluated as a different data-driven solution for the prediction of B(E2) values. The outcomes show that ensemble models, in particular GBMs, RF, and XGBoost, provide vastly improved predictive capabilities and generalizing influence while creating strong correlations to experimental data with small prediction errors. On the other hand, deep learning models such as CNN and NN is prone to overfitting, while simpler ones such as MLR and KNN fail to capture the non-linear relationships inherent in nuclear data. The findings underscore the promise of ensemble ML tools for nuclear physics in a scalable, accurate approach for predicting transition probabilities.