Résumé:
Uncertain pattern mining is considered as an NP-Hard problem due to its complexity and its
execution time consummation. The problem is amplified in the Big Data era. Thus, we need to use
techniques that don’t require prior knowledge of the search space as the metaheuristics algorithms,
which use natural theories based on randomness.
This work deals with the uncertainty of data when extracting frequent patterns from big
uncertain (probabilistic) Datasets (BDUPM for Big Data Uncertain Pattern Mining). In addition to
that, the BDUPM task is addressed as a combinatorial optimization problem in this study. In fact,
we proposed three metaheuristic-based algorithms that are inspired from the Particle Swarm
Optimization (PSO), Bee Swarm Optimization (BSO) and Genetic Algorithms (GA), for the
purpose of extracting unexpected useful frequent patterns that help to get useful pieces of
information to make trusted decisions.
The proposed algorithms MRPSO-UFIM, MRBSO-UFIM and MRGA-UFIM are employed
with the MapReduce programming model in a parallel and distributed environment, and examined
based on the number of frequent itemsets retrieved, and computational time. The experiments have
shown the efficiency of our proposed solutions when tested with several uncertain datasets.
Key words : Frequent Pattern Mining, Uncertain Data, Big Data, Particle Swarm Optimization, Genetic Algorithms, Bee Swarm Optimization.
