Trends of Evolutionary Machine Learning to Address Big Data Mining

Improving decisions by better mining the available data in an Information System is a common goal in many decision making environments. However, the complexity and the large size of the collected data in modern systems make this goal a challenge for mining methods. Evolutionary Data Mining Algorithms (EDMA), such as Genetic Programming (GP), are powerful meta-heuristics with an empirically proven efficiency on complex machine learning problems. They are expected to be applied to real-world big data tasks and applications in our daily life. Thus, they need, as all machine learning techniques, to be scaled to Big Data bases. This paper review some solutions that could be applied to help EDMA to deal with Big Data challenges. Two solutions are then selected and explained. The first one is based on the algorithmic manipulation involving the introduction of the active learning paradigm thanks to the active data sampling. The second is based on the processing manipulation involving horizontal scaling thanks to the processing distribution over networked nodes. This work explains how each solution is introduced to GP. As preliminary experiences, the extended GP is applied to solve two complex machine learning problem: the Higgs Boson classification problem and the Pulsar detection problem. Experimental results are then discussed and compared to value the efficiency of each solution.