Distance-based correlation analysis for graph databases

Abstract Big data is often characterized by its volume, velocity, and variety, properties that entail the fact that the data contains values and relationships that are too complex to be stored using standard, relational, or document databases. Graph databases, commonly utilized for their capacity to model complex relationships between sets of objects, provide an effective framework for the processing and storing of such data. Afterwards, it is necessary to work with data further − analyse it using methods of descriptive statistics and statistical analysis, visualize it with the use of exploratory analysis techniques, and especially use this data to build analytical models for predictive and estimation purposes. The main objective of the presented study is the design and implementation of the predictive potential metric in graph databases, which is based on the structures found in the graph databases themselves. We focus on the examination of the correlation between the attribute values of individual database objects and the mutual distance of these objects in the defined graph space. The proposed metric is verified using standard prediction models built on a sizeable graph database.