Investigation of Influence of Identification Precision on Modeling Accuracy in Belief Rule Base

Many current Belief Rule Base (BRB)-related training and learning studies have been focused on BRB parameter, BRB structure learning, or their combination. Comparatively, the number and utilities of scales have not been optimized mainly because they do not have direct influence on the structure of BRB. Nonetheless, it is not clear how the number and utilities of scales would affect the modeling accuracy. Therefore, two hypotheses are tested in this study, namely, whether including the utilities of scales would affect the modeling accuracy, and whether the number of scales would affect the modeling accuracy. With a practical pipeline leak detection case, it is validated that: (1) the utilities of scales do not have direct influence on the modeling accuracy, and (2) the number of scales do not have direct influence on the modeling accuracy. Further t-tests are conducted, which also produces consistent results.