Predictive analytics that takes in account network relations: A case study of research data of a contemporary university

Nankani, E. and Simoff, S.

    Contemporary organisations incorporate large amount of invisible networks between their employees. The structure of such networks impacts the information fusion within the organisation. Taking into account the inuence of such network structures in predictive modelling will be beneficial for the quality of organisational strategic planning. Network mining methods (the social network analysis of large heterogeneous data sets) can extract information about the structure of such networks and the strategic positioning of each individual from various interaction data. We propose to integrate the output of network mining into the predictive modelling cycle in order to depict these inuences. This paper demonstrates such approach by incorporating network centrality measures of actor closeness and actor betweeness in CART predictive modelling cycle. It presents a proof-of-concept application of this integrated approach to the case study of a contemporary university, which resembles some similarity with corporate organisations. The study utilises a data set about academic research activities collected over five years. The results of the study support the hypothesis that information about the network structures in a data set (whose impact is included through the centrality measures) can improve the accuracy of predictive analysis.
Cite as: Nankani, E. and Simoff, S. (2009). Predictive analytics that takes in account network relations: A case study of research data of a contemporary university. In Proc. Australasian Data Mining Conference (AusDM'09) Melbourne, Australia. CRPIT, 101. Kennedy P. J., Ong K. and Christen P. Eds., ACS. 99-108
pdf (from crpit.com) pdf (local if available) BibTeX EndNote GS