GEML: A Grammatical Evolution, Machine Learning Approach to Multi-class Classification
In this paper, we propose a hybrid approach to solving multi-class problems which combines evolutionary computation with elements of traditional machine learning. The method, Grammatical Evolution Machine Learning (GEML) adapts machine learning concepts f
- PDF / 676,965 Bytes
- 22 Pages / 439.37 x 666.142 pts Page_size
- 47 Downloads / 159 Views
Abstract. In this paper, we propose a hybrid approach to solving multiclass problems which combines evolutionary computation with elements of traditional machine learning. The method, Grammatical Evolution Machine Learning (GEML) adapts machine learning concepts from decision tree learning and clustering methods and integrates these into a Grammatical Evolution framework. We investigate the effectiveness of GEML on several supervised, semi-supervised and unsupervised multiclass problems and demonstrate its competitive performance when compared with several well known machine learning algorithms. The GEML framework evolves human readable solutions which provide an explanation of the logic behind its classification decisions, offering a significant advantage over existing paradigms for unsupervised and semi-supervised learning. In addition we also examine the possibility of improving the performance of the algorithm through the application of several ensemble techniques. Keywords: Multi-class classification · Grammatical evolution tionary computation · Machine learning
1
· Evolu-
Introduction
Evolutionary algorithms (EAs) are algorithms which are inspired by biological evolution and which are constructed to emulate aspects of evolution, such as genetic mutation and recombination and the notion of natural selection. Genetic Programming (GP) [29] is an evolutionary algorithm which has been successful on a wide range of problems from various diverse domains [19], achieving many human competitive results [4]. However, a significant proportion of previous work has concentrated on supervised learning tasks and, aside from some notable exceptions, studies on unsupervised and semi-supervised learning have been left to the wider machine learning (ML) community. Two of the most important problems types which benefit from the application of ML techniques are regression and classification, and GP has proven itself as an effective learner on each of these: achieving particularly competitive results on symbolic regression and binary classification tasks. Although many studies have c Springer International Publishing AG 2017 J.J. Merelo et al. (eds.), Computational Intelligence, Studies in Computational Intelligence 669, DOI 10.1007/978-3-319-48506-5 7
114
J.M. Fitzgerald et al.
been undertaken, multi-class classification (MCC) remains a problem which is considered challenging for traditional tree based GP [11]. While we are concerned with multi-class classification generally, an important motivation for the current investigation is the requirement for an algorithm which can be applied to multi-class grouping/categorisation tasks involving both labelled and unlabelled inputs from the medical domain, where the unsupervised algorithm must be able to supply human interpretable justification for categorisation decisions. Clustering is a natural choice for this type of task, but standard clustering algorithms generally fail to satisfy the requirement of providing the reasoning behind cluster allocations in a human readable form. In the med
Data Loading...
