Genetic Programming for Machine Learning
Alain Pétrowski, Sana Ben-Hamida · 2017
Genetic programming (GP) is considered as the evolutionary technique having the widest range of application domains. It can be used to solve problems in at least three main fields: optimization, automatic programming and machine learning. This chapter summarizes the different GP implementations based on one of the three representations: tree-based representation, linear-based representation and graph-based representation. It presents three of these implementations that have proven successful in practice: linear GP (LGP), grammatical evolution (GE) for linear-based representation and Cartesian GP (CGP) for graph-based representation. Several research papers explore the feasibility of applying GP to multi-category pattern classification problems. The chapter proposes a CGP-based approach to design classifiers for an Intrusion Detection problem. The major problem faced by an intrusion detection system (IDS) is the large number of false-positive alerts, i.e. normal behaviors mistakenly classified as alerts.