A Novel Evolutionary Automatic Clustering Technique by Unifying Initial Seed Selection Algorithms into Teaching–Learning-Based Optimization

This paper endeavors to embark upon one of the key inadequacies of simple k-means partitioning clustering algorithm; i.e., the number of clusters is precise and is yet to be initialized before the algorithmic execution, and the algorithms confinement of squared error function is toward local optima. This restriction may even affect to bestow on overlapping partitions in the given dataset; the anticipation from this premeditated work is to find automatically the optimal number of clusters, endorse the incurred clusters with cluster validity indices (CVIs), and eventually estimate the minimum consumption in percentage of error rate and CPU time when enforced over real-time datasets. This expectancy is down to earth with a unified approach entrenched into teaching–learning-based optimization (TLBO) by constituting an initial seed selection strategic algorithm at initialization step, thereby clusters configuration and affirmation in teacher and learner phases. Experimental results substantiate that this clustering framework efficaciously tackles the aforementioned limitations and capitulate promising outcomes.