Optimization of K-means algorithm: Ant colony optimization

Significance of a versatile and simple clustering algorithm is becoming indispensable with the huge data growth in recent years. K-Means clustering is one such clustering algorithm which is simple yet elegant. But K-Means Algorithm has its disadvantages, dependence on the initial cluster centers and the algorithm tends to converge at a local minima. To overcome these disadvantages, ant colony optimization is applied to improve the traditional K-Means clustering algorithm. Two methods of using ants in K-Means are presented in the paper. In the first method the ant is allowed to go for a random walk and picks a data item. Pick and Drop probabilities of that particular data item are calculated. These values determine whether a data item remains in the same cluster or is moved to another cluster. In the second method instead of letting the ant pick up a data item randomly we calculate the pick and drop and let the ant walk to the data item which has the highest probability to be moved to another cluster. Entropy and F-measure are considered as quality measures.