Managing uncertainty during CBR systems vocabulary maintenance using Relational Evidential C-Means

Due to the incremental learning of Case-Based Reasoning (CBR) systems, there is a colossal need to maintain their knowledge containers which are (1) the case base, (2) similarity measures, (3) adaptation, and (4) vocabulary knowledge. Actually, the vocabulary presents the basis of all the other knowledge containers since it is used for their description. Besides, CBR systems store real-world experiences which are full of uncertainty and imprecision. Therefore, we propose, in this paper, a new policy to maintain vocabulary knowledge using one of the most powerful tools for uncertainty management called the belief function theory, as well as the machine learning technique called Relational Evidential C-Means (RECM). We restrict the vocabulary knowledge to be the set of features describing cases, and we aim to eliminate noisy and redundant attributes by taking into account the correlation between them.