Belief Discernibility Matrix and Function for Incremental or Large Data

Belief functions and rough sets: Survey and new insights

“…The definition of the upper approximation is unchanged. As we show in Section 4.3, the notion of UDT and the above definitions of approximations have been used by Lingras et al [93,95,96] to perform feature selection and classification based on uncertain (in particular, evidential) data. -…”

“…In particular, to search for the reducts of an UDT (and hence perform feature selection), the same authors provide a heuristic algorithm based on a generalized definition of the discernibility matrix [95]. In [96,97], they also propose a parallel algorithm for application to big data.…”

“…Remarks and Prospects. While the authors of [96] studied the performance of the discussed feature reduction and classification methods and reported promising empirical results, no comparison with other classifiers for evidential data [5,20,32,40,73] has so far been evaluated in the literature. Furthermore, similarly to the remarks in Section 3.3, also for the case of feature reduction and rule induction it would be interesting to evaluate the behavior and performance of different aggregation as well decision rules.…”

“…Furthermore, similarly to the remarks in Section 3.3, also for the case of feature reduction and rule induction it would be interesting to evaluate the behavior and performance of different aggregation as well decision rules. In respect to this latter aspect, while the authors of [96] considered only the pignistic transform, other decisionmaking criteria such as mentioned in Section 4.1 or disambiguation methods [42] should also be evaluated. A final remark regards the relationships among the methods for feature selection proposed in [93] and [13].…”