Multiobjective evolutionary approaches to pattern classification and robust optimization

Created by W.Langdon from gp-bibliography.bib Revision:1.9225

@PhdThesis{Kaustuv_Nag:thesis,

• author = "Kaustuv Nag",
• title = "Multiobjective evolutionary approaches to pattern classification and robust optimization",
• school = "Department of Instrumentation and Electronics Engineering, Jadavpur University, Kolkata",
• year = "2018",
• address = "Kolkata, India",
• keywords = "genetic algorithms, genetic programming, Multiobjective optimization problem, Pattern Recognition, Robust optimization",
• language = "English",
• URL = "https://irju.jdvu.ac.in/bitstream/123456789/1029/1/PhD%20thesis%20%28I.E.E.%29%20Kaustuv%20Nag.pdf",
• URL = "http://hdl.handle.net/10603/356239",
• URL = "https://shodhganga.inflibnet.ac.in/handle/10603/356239",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/1/01_title.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/2/02_dedication.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/3/03_university%20submission%20form.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/4/04_certificate.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/5/05_acknowledgements.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/6/06_contents.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/7/07_list%20of%20figure%2c%20table%20%26%20algorithms.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/8/08_chapter%201.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/9/09_chapter%202.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/10/10_chapter%203.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/11/11_chapter%204.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/12/12_chapter%205.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/13/13_chapter%206.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/14/14_appendix.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/15/15_bibliography.pdf",
• URL = "https://shodhganga.inflibnet.ac.in/bitstream/10603/356239/16/80_recommendation.pdf",
• size = "xxxi+175 pages",
• abstract = "In this thesis, we use multiobjective evolutionary approaches to address four research problems related to decision making. First, we propose a method for simultaneous feature selection (FS) and classification. In the next work, we propose a method of feature extraction and selection for designing parsimonious classifiers. In the succeeding work, in the context of fuzzy group decision making for multiobjective optimization problems, we propose a new measure, called robust consensus, and incorporate it in the search process. Finally, we propose a measure of sensitivity that can assess the sensitivity, and hence, robustness of a multiobjective solution. In the same work, we also propose three approaches to use this measure in the search process.

  In the first work, we present an integrated algorithm for simultaneous FS and design of diverse classifiers using a steady state multiobjective genetic programming, which minimizes three conflicting objectives: false positives, false negatives, and the number of leaf nodes in the tree. Our method divides a c-class problem into c binary classification problems. It evolves c sets of genetic programs to create c ensembles. During mutation operation, our method exploits the fitness as well as the unfitness of features, which dynamically change with generations with a view to using a set of highly relevant features with low redundancy. The classifiers of the i th class determine the net belongingness of an unknown data point to the i th class using a weighted voting scheme, which makes use of the false positive and false negative mistakes made on the training data. We test our method on eight microarray and eleven text data sets with a diverse number of classes (from 2 to 44), a large number of features (from 2000 to 49151), and a high feature-to-sample ratio (from 1.03 to 273.1). We compare our method with a bi-objective genetic programming scheme that does not use any FS and rule size reduction strategy. It helps to depict the effectiveness of the proposed FS and rule size reduction schemes. Furthermore, we compare our method with four classification methods in conjunction with six features selection algorithms and the full feature set. Our scheme performs the best for 380 out of 474 combinations of data sets, classification algorithm and FS method.

  The primary objective of the next work is to design classifiers that are reasonably simple and understandable yet capable of yielding a good performance. Thus, our problem here is to obtain an ensemble of linear classifiers with as few simple features as possible. For this we use genetic programming to extract features that can linearly separate two classes. We propose an integrated mechanism for simultaneously extracting and selecting useful linearly separable features. We decompose a c-class problem into c binary classification problems and evolve c sets of binary classifiers employing a steady-state multiobjective genetic programming with three minimizing objectives. Each binary classifier is composed of a binary tree and a linear support vector machine (SVM). The features extracted by the feature nodes and some of the function nodes of the tree are used to train the SVM. The decision made by the SVM is considered the decision of the corresponding classifier. During crossover and mutation, the SVM-weights are used to determine the usefulness of the corresponding nodes. We also employ unfitness functions for the feature nodes and a fitness function based on Golub index to select useful features. We compare our method with 34 classification systems using 18 data sets. The performance of the proposed method is found to be better than 432 out of 570, i.e., 75.79 percent of comparing cases. Our results confirm that the proposed method is capable of achieving our objectives.

  In the succeeding work, we consider a multiobjective robust optimization problem where a set of weighted decision makers provides their preferences a priori. The preferences are provided either in the objective space or in the decision variable space using fuzzy numbers. To solve this problem the following three things are required: 1) an indicator to measure consensus, 2) an indicator to measure the robustness of the solutions to their degrees of consensus, and 3) a reformulation of the multiobjective robust optimization problem. It is necessary for the reformulated problem to generate robust solutions that also enjoy high degrees of consensus. In this work, we have addressed these three issues. For this purpose, we have proposed two approaches to define consensus. Then, we have extended these approaches to define robust consensus, an indicator to measure the robustness of a given solution to its degree of consensus. Though these approaches can be used to define a countless number of measures, we have proposed 12 definitions of consensus, and hence, robust consensus. Furthermore, we have proposed two ways for the reformulation. Experimental results illustrate that the behaviors of the proposed definitions and of the reformulations are consistent with our expectations.

  In the final work, in the context of multiobjective optimization problems, we have introduced a new measure of sensitivity, which is inversely related to robustness of solutions. It quantifies the robustness of a solution with respect to perturbations in the variable space. We have shown how the cost of computing this measure can be reduced using an approximation with the first-order Taylor series expansion subject to the following three conditions. First, the objective functions are differentiable. Second, the input noise BibTeX entry too long. Truncated

Genetic Programming entries for Kaustuv Nag

Citations