Abstract:

Traditional kmeans clustering (KMC) algorithm is overdependent on initial value setting and falls into local optimum easily. Shuffled frog leaping algorithm (SFLA) has some shortcomings, such as slow speed on convergence and searching, incomprehensive exchange between local and global information. Aiming at these disadvantages, we propose a modified shuffled frog leaping algorithm (MSFLA). According to the ideas of differential evolution and particle swarm optimization, inertia weight coefficients of former displacement and scaling factors are introduced into the MSFLA during individual variation of frogs. We randomly choose a point between the best location and the best historical position，and take the difference value between the average and the worst position as the step length to update individual frogs . We present the MSFLAKMC based on the MSFLA and the KMC, which effectively overcomes the problems of initial value setting of the KMC algorithm, and reduces the likelihood of the KMC algorithm into a local optimum. Experimental results show that the MSFLA has strong search capabilities while the MSFLAKMC has better clustering performance.

Key words: Kmeans clustering algorithm;SFLA;distance updating formula;fitness function;clustering