| Input: Information table IS = (U, A, V, f), where U = {x1, ..., xn}, A = {a1, ..., am}; the number of clusters k expected. |
| Output: k initial center points. |
| Input: Information table IS = (U, A, V, f), where U = {x1, ..., xn}, A = {a1, ..., am}; the number of clusters k expected. |
| Output: k initial center points. |
| Initialization: Let C = Φ, where C is the initial set of center points that have been selected |
(1) | Calculate the division U/IND (A-{a}) and U/IND ({a}) respectively by counting sorting; |
(1.1) | Calculate the information entropy E (A-{a}) of IND (A-{a}); |
(1.2) | Calculate the importance of the attribute a Sig (a), and thus obtain the weight of a weight (a); |
(1.3) | For any x ∈ U, calculate |[x]{a}| according to the division U/IND ({a}), and |
(2) | Calculate WDens (x) for any x ∈ U; |
(3) | Select the object y with the largest weighted average density from U as the first initial center, and C = C{y}; |
(4) | If |C| < k, go to step (5), otherwise go to step (10); |
(5) | Assume that C = {c1, c2, ..., cq}, repeated for any x ∈ U -C |
(5.1) | Calculate the weighted overlapping distance wd (x, ci) of x and ci, where ci ∈ C, 1 ≤ i ≤ q; |
(5.2) | Calculate Pos_ Center (x); |
(6) | Select the object y that is the most likely to be the initial center from U-C as the new initial center. |
And let C = C{y}; |
(7) | If |C| < k, go to step (5), otherwise go to step (8); |
(8) | Return k initial centers in C. |