The major weakness of genetically guided clustering is the CPU time it requires. Clustering is generally slow and the GGA clustering approach is even slower. The dominant cost is that of calculating the fitness value for each population member for each generation. This cost is made up of the the number of features in a data set, the number of objects or feature vectors in a data set and the number of classes to cluster the data into. The number of bits needed to represent features has a lesser effect on the time per generation. GA speed could be much improved by a parallel implementation where the P fitness values are calculated in parallel for each generation.
The GA approach to clustering requires time proportional to the number of bits per feature, population size, number of features, number of feature vectors and number of clusters. An increase in any of these parameters results in a linear increase in time per generation for the GA clustering approach.
The Iris data set is as a good concrete example of the times required by the GA approach to clustering as opposed to FCM or HCM. Table 6 shows the time per generation or iteration in seconds for GGA on the HCM functional (GGAHCM), GGA applied to the FCM functional (GGAFCM), FCM and HCM on the Iris data. The GA population is size 50. HCM is about twice as fast as FCM. The GA approach is about 20 times slower than HCM/FCM for a generation vs. iteration. The maximum number of iterations for HCM/FCM to converge on the Iris data is 20, while the GA will need close to 300 generations. Hence, GGA clustering can take up to 2 orders of magnitude longer than FCM or HCM in some cases.
| GGAHCM | HCM | FCM | GGAFCM |
| 0.022 | 0.00111 | 0.00222 | 0.043 |