الفهرس Only 14 pages are availabe for public view
 |  | from | 110 |  |  |
| | | from | 110 | | |
Abstract
Some approaches have been introduced before to unsupervised fuzzy classification of multidimensional data where patterns are considered to belong to some but not necessarily all clusters. Such algorithms are alled ’semi-fuzzy’ or ’soft’ clustering techniques. This thesis introduces some new modifications to the soft clustering techniques that increase the speed of convergence while keeping the performance levels, measured by the objective function, as they are. The proposed ASCM clustering algorithm lets each ordinary iteration of the soft clustering be followed by an improvement stage. Once tflt cluster centre locations are updated by the regular soft clustering algorithm operations, the improvement stage shifts each cluster centre farther in its respective update direction. Another proposed technique, which is Gauss-Seidel Like clustering algorithm is also proposed to improve the speed of the soft clustering algorithm. Instead of calculating the locations of the new centres after an iteration that calculates’ the membership values of all patterns, it updates the locations of the centres after each pattern’s membership is calculated. This thesis also proposes another variation of the . SCPM (soft clustering with proportional membership) algorithm by making an initial iteration of the FCM (fuzzy c-means) to initialize the values of membership; which considerably accelerates the SCPM