Drill-CODA: A Framework for Supporting Drill-Across Multidimensional Big Data Analytics Over Big Co-Occurrence Aggregate Hierarchical Data
摘要
This paper proposes and experimentally assesses the Drill-CODA framework, which allows us to support drill-across multidimensional big data analytics over big co-occurrence aggregate hierarchical data, with also privacy-preservation features. Drill-CODA is a composite framework that combines several data processing and analytics metaphors over hierarchical data, all in a multidimensional fashion, with the goal of providing useful insights over large-scale big data repositories, while protecting their privacy. The latter is, as for now, a critical challenge in the big data research community, which arises in a plethora of emerging big data application scenarios, ranging from urban analytics to social network analysis, from bio-medical tools to industry 4.0 prognostic tools, and so forth. This is because data from real-life settings are hierarchical by nature. To validate its effectiveness, we conducted three experimental evaluations using six real-life public health datasets, including Mental Disorders, C15 Plus, Substance Use, Narcan Administration, Diabetes, and Cancer Deaths. The results highlight the framework ability to uncover strong correlations across heterogeneous data domains: for example, Pearson correlation values reached up to 0.92 (with corresponding Spearman coefficients around 0.89) in mental disorder/substance use analysis, while diabetes/cancer mortality correlations ranged from 0.80 to 1.00 across countries such as Italy, Germany, and France. Furthermore, the cancer incidence/mental disorder analysis revealed heterogeneous patterns, with South/Central America exhibiting strong correlations (