A Novel Algorithm for Evaluating Clustering Propensity of IoT-Generated Spatio-Temporal Data Geared for Distributed Systems

Spatio-temporal (ST) data generated by Internet of Things (IoT) devices is expected to grow exponentially in the future on a massive scale. The clustering of this massive amount of space-time data is expected to help understand the underlying characteristics of the data-generating phenomenon for further analysis. However, most clustering algorithms require the data to be in a central location for effective cluster detection. Therefore, dealing with this massive amount of ST data in terms of communication overhead and the capacity of the central server to store and analyze it to generate actionable knowledge for real-time applications would be challenging. This article proposes a novel algorithm, enhanced noncontiguous to contiguous visual assessment of clustering tendency (enccVAT), geared toward distributive computing and extracts clusters with the highly desired space-time contiguity property from the massive amount of IoT-generated ST data. Furthermore, the proposed algorithm can effectively estimate the number of clusters in the dataset before clustering (the clustering tendency assessment problem), which is usually required by comparable algorithms. We performed experiments on five large ST datasets to validate the effectiveness of the proposed scheme compared to three state-of-the-art algorithms. The results expressed in terms of CPU run time of algorithms, Dunn’s index (DI), Silhouette indices, and four approximations of DI showcase the superiority of the proposed solution over comparable approaches.