Fig. 3.

Dimensional selection and evaluation. (A) Using FCM paired with MPC, the number of potential clusters is evaluated. The dimension's dataset is sorted into different numbers of clusters, and then MPC chooses the optimal number of clusters. (B) If all dimensions are unimodal, the recursive clustering process ends. (C) If at least one dimension is multimodal, the space is constructed from the dimensions that have two or more modes (Upper). All unimodal dimensions are then removed from the space (Lower). (D) Each dimension is examined for number of modes (“C”). Dimensional importance (DI) is calculated for each of the dimensions. Feature space is transformed by weighting (scaling) each dimension by dimensional importance.