A generalized automatic clustering algorithm using improved TLBO framework

Many clustering algorithms have been proposed, yet most of them require predefined number of clusters. Unfortunately, unavailable information regarding number of clusters is commonly happened in real-world data of different domains. This study is aimed to overcome the above stated problem by developing a generalized automatic clustering algorithmic framework (Auto ITLBO) by combining the self-acting initial seeding algorithm (SPSS) into improved Teaching-Learning-Based Optimization (TLBO)algorithm to partition the data automatically into appropriate number of clusters.

This generalized automatic clustering framework possess multi-objectives to automatically evolve proper number of optimal partitions, clustering accuracy with minimum error rate and CPU time, and validating the prompted clusters using cluster validity indices (CVIs). These multiple objectives are optimized simultaneously using improved TLBO to determine a single set of solutions at the end of each run. The effectiveness of the proposed algorithm is compared with other popular meta-heuristic techniques and is extensively verified over artificial and real-time data sets of varying complexities. Results of experiments demonstrate that proposed algorithm is more efficient in attaining the assumed objectives. Thus, the feasibility of our generalized automatic clustering framework was validated.