SRIHASS - a similarity measure for discovery of hidden time profiled temporal associations

• Published in: Multimedia tools and applications 2018-07, Vol.77 (14), p.17643-17692
• Main Authors: Radhakrishna, Vangipuram, Veereswara Kumar, Puligadda, Janaki, Vinjamuri
• Format: Article
• Language: English
• Subjects: Comparative analysis; Computer Communication Networks; Computer Science; Data mining; Data Structures and Information Theory; Distance measurement; Euclidean geometry; Multimedia Information Systems; Normal distribution; Similarity; Similarity measures; Special Purpose and Application-Based Systems; Standard scores; Statistical analysis
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-017-5185-9

Mining and visualization of time profiled temporal associations is an important research problem that is not addressed in a wider perspective and is understudied. Visual analysis of time profiled temporal associations helps to better understand hidden seasonal, emerging, and diminishing temporal trends. The pioneering work by Yoo and Shashi Sekhar termed as SPAMINE applied the Euclidean distance measure. Following their research, subsequent studies were only restricted to the use of Euclidean distance. However, with an increase in the number of time slots, the dimensionality of a prevalence time sequence of temporal association, also increases, and this high dimensionality makes the Euclidean distance not suitable for the higher dimensions. Some of our previous studies, proposed Gaussian based dissimilarity measures and prevalence estimation approaches to discover time profiled temporal associations. To the best of our knowledge, there is no research that has addressed a similarity measure which is based on the standard score and normal probability to find the similarity between temporal patterns in z-space and retains monotonicity. Our research is pioneering work in this direction. This research has three contributions. First, we introduce a novel similarity (or dissimilarity) measure, SRIHASS to find the similarity between temporal associations. The basic idea behind the design of dissimilarity measure is to transform support values of temporal associations onto z-space and then obtain probability sequences of temporal associations using a normal distribution chart. The dissimilarity measure uses these probability sequences to estimate the similarity between patterns in z-space. The second contribution is the prevalence bound estimation approach. Finally, we give the algorithm for time profiled associating mining called Z-SPAMINE that is primarily inspired from SPAMINE. Experiment results prove that our approach, Z-SPAMINE is computationally more efficient and scalable compared to existing approaches such as Naïve, Sequential and SPAMINE that applies the Euclidean distance.