Lecture Notes in Computer Science, Volume 7833, Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics: 11th European Conference, EvoBIO 2013, Vienna, Austria, April 3-5, 2013. Proceedings
Format: PDF / Kindle (mobi) / ePub
This book constitutes the refereed proceedings of the 11th European Conference on Evolutionary Computation, Machine Learning and Data Mining in Bioinformatics, EvoBIO 2013, held in Vienna, Austria, in April 2013, colocated with the Evo* 2013 events EuroGP, EvoCOP, EvoMUSART and EvoApplications. The 10 revised full papers presented together with 9 poster papers were carefully reviewed and selected from numerous submissions. The papers cover a wide range of topics in the field of biological data analysis and computational biology. They address important problems in biology, from the molecular and genomic dimension to the individual and population level, often drawing inspiration from biological systems in oder to produce solutions to biological problems.
and 2.2 (the case is also passed to the classiﬁer at node 1-2.1 if it is classiﬁed as belonging to class 1 and 2) at the next level. If the case does not belong to class 2, no further classiﬁcations are performed from node 2. In this example, a case can be labelled with, say, classes 1, 2, and 2.2, which is appropriate for the aforementioned characteristics of our dataset. 86 5 K.M. Salama and A.A. Freitas Proposed Methods for an Ensemble of Classiﬁers An ensemble of classiﬁers is often
Statistical Epistasis Network by Hu et al, the threshold of including pairwise interactions can be derived systematically by analyzing the topological properties of the networks , such as the size of a network, the connectivity of a network (the size of its largest connected component), and its vertex degree distribution. Permutation testing is often used to provide a null distribution of properties of networks built from permuted data. This null distribution can be used to determine the
and the F-score values of the stress classifications derived from the classifier, its parameters and a subset of stress features encoded by the chromosome. The classifiers used for stress classification were ANNs and SVMs. An ANN, inspired by biological neural networks, has the capability to learn patterns to recognize characteristics in input tuples by classes. It is made up of interconnected processing elements, known as artificial neurons, which are connected by weighted links that pass
CGPANN algorithm is positive and encouraging, outperforming a number of other published approaches as referenced in Table 1, and ameliorating CGPANN performance on the problem while simultaneously improving eﬃciency. The results demonstrate that through our modiﬁcations, CGPANN is capable of successfully producing RBF networks. Improving CGPANN Performance 173 Table 2. Performance of standard CGPANN and our modiﬁed approach in terms of correctly classiﬁed records (CCR) and a breakdown of
incorrectly classiﬁed 8 exemplars. One of these exemplars was randomly selected and subjected to further informal evaluations using the CGPANN networks, various commercially available neural network packages and the MFF-NEAT algorithm, with varying parameters and distributions of the training and testing data. All trialled approaches failed to correctly classify this exemplar. This does not necessarily suggest that this exemplar is incorrectly labelled, but perhaps the reasoning behind its