AGBWTDRS-EM: Multi-criteria decision-making approach using adaptive granular-ball weighted temporal rough set and ensemble learning

Advances in technology and society have introduced new characteristics to decision-making problems, such as multi-criteria, multi-granularity, temporality, and incompleteness, which exceed the scope of traditional decision-making methods. This paper proposes a multi-criteria uncertain decision-making method (AGBWTDRS-EM), which integrates adaptive granular-ball weighted temporal rough set with ensemble learning, centred on rough set theory and dominance relations. First, a temporal hybrid information system is constructed, with preprocessing and normalisation methods for different attribute types. Then, temporal binary relations are defined in the temporal hybrid information system, and granular balls generated via K-means clustering form the granular-ball temporal rough set. Attribute weights are calculated based on the dependence of the condition attributes on the decision attributes, and a parameter adaptive mechanism constructs the adaptive granular-ball weighted temporal rough set (AGBWTDRS); its mathematical properties and degradation conditions are discussed. Finally, AGBWTDRS is fused with an ensemble learning model of nine base classifiers tailored to distinct data characteristics, establishing the AGBWTDRS-EM uncertainty decision model. The model achieved mean accuracy, precision, recall, and F1 scores of 0.8945, 0.9240, 0.8945, and 0.8884, outperforming all compared models on 18 out of 20 datasets. Sensitivity analysis and statistical tests confirmed its robustness and superiority. This study extends the real-world applications of rough-set theory and machine learning and builds an organic correlation and integration mechanism between the traditional uncertain decision-making methods and the new methods for current uncertain decision-making problems. All the codes were released at https://github.com/zxx-asd/AGBWTDRS-EM.