An Efficient Pre-Clusters Assessment Technique for Efficient Data Partitions

Massive information is being created over smartphones, IoT devices, and social media. Clusters are necessary for managing such unlabeled massive data for organizing the vast data into manageable data partitions. The survey is conducted on visual clustering methods for determining the number of clusters (also known as cluster tendency). These algorithms visually displayed the number of clusters as square-shaped dark blocks along the diagonal. Cluster tendency is depicted as a series of square, black blocks. The techniques, as mentioned earlier, perform admirably on small datasets. However, the clustering accuracy diminishes on large datasets and processing time increases. This paper proposed a system that improves upon the current state of the art in visual approaches to address this issue. Instead of using the traditional Euclidean distance to evaluate clustering, this work proposes a cosine metric for their visual algorithms. Experiments illustrating and contrasting the efficacy of proposed and existing visual clustering algorithms are carried out on benchmarked large datasets.