|
| Clustering method | Analysis | Limitation |
|
| k-prototypes [10] | Step 1. Select several data points as the initial cluster centres randomly | Data can only be allocated into one cluster, losing information about multiple membership;
Sensitive to initialization of cluster centres;
No discretion of varying weights of an attribute in different clusters |
| Step 2. Allocate the rest data points with the highest similarity to the relevant cluster centre |
| Step 3. Update the cluster centres after each data allocation |
| Step 4. Recalculate the dissimilarity of each data point to each cluster centre |
| Step 5. Repeat steps 2 to 4 until reaching the optimal cluster allocation |
|
| SBAC [11] | Step 1. Create a cluster that contains a pair of data points with the highest similarity | Uses the same method to measure both numerical and categorical data;
Sensitive to initialization of cluster centres;
No discretion of varying weight of an attribute in different clusters |
| Step 2. Add another data point and compare its similarity with the initial two data points. If the similarity is higher than that of the two original ones, add to the cluster; otherwise treat the data as a new cluster centre |
| Step 3. Repeat Step 2 until all data points have been located to a cluster |
| KL-FCM-GM [13] | Step 1. Use gath-geva theory to allocate fuzzy membership of all data points | Sensitive to initialization of fuzzy membership |
| Step 2. Update weight of each data point in the allocated clusters |
| Step 3. Retest the distance of data point to cluster centre |
| Step 4. Estimate the cluster distribution parameters of gath-geva formula again |
| Step 5. Repeat steps 1 to 4 until the objective function has reached the stop condition |
|
| IWKM [14] | Step 1. Select several data points as the initial cluster centres randomly and assign weight to each | Sensitive to initialization of cluster centres |
| Step 2. Compute fuzzy membership matrix based on the values of the initial cluster centres and their weights obtained in Step 1 |
| Step 3. Update cluster centres with the values of fuzzy membership matrix and weights in Step 2 |
| Step 4. Update the weights with the values of fuzzy membership matrix and cluster centres in Step 3 |
| Step 5. Repeat steps 2 to 4 until the objective function has reached the stop condition |
|
