Genetic Programming over Spark for Higgs Boson Classification

With the growing number of available databases having a very large number of records, existing knowledge discovery tools need to be adapted to this shift and new tools need to be created. Genetic Programming (GP) has been proven as an efficient algorithm in particular for classification problems. Notwithstanding, GP is impaired with its computing cost that is more acute with large datasets. This paper, presents how an existing GP implementation (DEAP) can be adapted by distributing evaluations on a Spark cluster. Then, an additional sampling step is applied to fit tiny clusters. Experiments are accomplished on Higgs Boson classification with different settings. They show the benefits of using Spark as parallelization technology for GP.