"Efficient Aggregation Methods for Probabilistic Data Streams"
: Proceedings of 8th International Symposium Business Modeling and Software Design (BMSD), Serie LNBIP, Vol. 319, 7-2018
Efficient Aggregation Methods for Probabilistic Data Streams
Sprache des Titels:
Proceedings of 8th International Symposium Business Modeling and Software Design (BMSD)
In this paper, we consider aggregation algorithms for SUM operator for uncertain stream processing. Deterministic algorithms can not be used here because of uncertain data and high rates of data change, time and memory constraints. We compare the most promising available methods. Instead of full distribution functions of query result, we use a set of six parameters based on key moments and quantiles to describe the distributions. It enables us to perform fast recomputations of the agregation with O(1) complexity. Experimental results demonstrate good performance of uncertain aggregation in comparison to deterministic case. We also found that usage of central limit theorem may be restricted to problems where data satisfy certain conditions.