Big Data is a new field, with many technological challenges to be understood in order to use it to its full potential. These challenges arise at all stages of working with Big Data, beginning with data generation and acquisition. The storage and management phase presents two critical challenges: infrastructure, for storage and transportation, and conceptual models. Finally, to extract meaning from Big Data requires complex analysis. Here the authors propose using metaheuristics as a solution to these challenges; they are first able to deal with large size problems and secondly flexible and…mehr
Big Data is a new field, with many technological challenges to be understood in order to use it to its full potential. These challenges arise at all stages of working with Big Data, beginning with data generation and acquisition. The storage and management phase presents two critical challenges: infrastructure, for storage and transportation, and conceptual models. Finally, to extract meaning from Big Data requires complex analysis. Here the authors propose using metaheuristics as a solution to these challenges; they are first able to deal with large size problems and secondly flexible and therefore easily adaptable to different types of data and different contexts. The use of metaheuristics to overcome some of these data mining challenges is introduced and justified in the first part of the book, alongside a specific protocol for the performance evaluation of algorithms. An introduction to metaheuristics follows. The second part of the book details a number of data mining tasks, including clustering, association rules, supervised classification and feature selection, before explaining how metaheuristics can be used to deal with them. This book is designed to be self-contained, so that readers can understand all of the concepts discussed within it, and to provide an overview of recent applications of metaheuristics to knowledge discovery problems in the context of Big Data.Hinweis: Dieser Artikel kann nur an eine deutsche Lieferadresse ausgeliefert werden.
Clarisse DHAENENS is Professor at the University of Lille in France and belongs to a research team working with both CRIStAL Laboratory (UMR CNRS) and Inria. Laetitia JOURDAN is Professor at the University of Lille in France and belongs to a research team working with both CRIStAL Laboratory (UMR CNRS) and Inria.
Inhaltsangabe
Acknowledgments xi Introduction xiii Chapter 1 Optimization and Big Data 1 1.1 Context of Big Data 1 1.1.1 Examples of situations 2 1.1.2 Definitions 3 1.1.3 Big Data challenges 5 1.1.4 Metaheuristics and Big Data 8 1.2 Knowledge discovery in Big Data 10 1.2.1 Data mining versus knowledge discovery 10 1.2.2 Main data mining tasks 12 1.2.3 Data mining tasks as optimization problems 16 1.3 Performance analysis of data mining algorithms 17 1.3.1 Context 17 1.3.2 Evaluation among one or several dataset(s) 18 1.3.3 Repositories and datasets 20 1.4 Conclusion 21 Chapter 2 Metaheuristics - A Short Introduction 23 2.1 Introduction 24 2.1.1 Combinatorial optimization problems 24 2.1.2 Solving a combinatorial optimization problem 25 2.1.3 Main types of optimization methods 25 2.2 Common concepts of metaheuristics 26 2.2.1 Representation/encoding 27 2.2.2 Constraint satisfaction 28 2.2.3 Optimization criterion/objective function 28 2.2.4 Performance analysis 29 2.3 Single solution-based/local search methods 31 2.3.1 Neighborhood of a solution 31 2.3.2 Hill climbing algorithm 33 2.3.3 Tabu Search 34 2.3.4 Simulated annealing and threshold acceptance approach 35 2.3.5 Combining local search approaches 36 2.4 Population-based metaheuristics 38 2.4.1 Evolutionary computation 38 2.4.2 Swarm intelligence 41 2.5 Multi-objective metaheuristics 43 2.5.1 Basic notions in multi-objective optimization 44 2.5.2 Multi-objective optimization using metaheuristics 47 2.5.3 Performance assessment in multi-objective optimization 51 2.6 Conclusion 52 Chapter 3 Metaheuristics and Parallel Optimization 53 3.1 Parallelism 53 3.1.1 Bit-level 53 3.1.2 Instruction-level parallelism 54 3.1.3 Task and data parallelism 54 3.2 Parallel metaheuristics 55 3.2.1 General concepts 55 3.2.2 Parallel single solution-based metaheuristics 55 3.2.3 Parallel population-based metaheuristics 57 3.3 Infrastructure and technologies for parallel metaheuristics 57 3.3.1 Distributed model 57 3.3.2 Hardware model 58 3.4 Quality measures 60 3.4.1 Speedup 60 3.4.2 Efficiency 61 3.4.3 Serial fraction 61 3.5 Conclusion 61 Chapter 4 Metaheuristics and Clustering 63 4.1 Task description 63 4.1.1 Partitioning methods 65 4.1.2 Hierarchical methods 66 4.1.3 Grid-based methods 67 4.1.4 Density-based methods 67 4.2 Big Data and clustering 68 4.3 Optimization model 68 4.3.1 A combinatorial problem 69 4.3.2 Quality measures 69 4.3.3 Representation 76 4.4 Overview of methods 81 4.5 Validation 82 4.5.1 Internal validation 84 4.5.2 External validation 84 4.6 Conclusion 86 Chapter 5 Metaheuristics and Association Rules 87 5.1 Task description and classical approaches 88 5.1.1 Initial problem 88 5.1.2 A priori algorithm 89 5.2 Optimization model 90 5.2.1 A combinatorial problem 90 5.2.2 Quality measures 90 5.2.3 A mono- or a multi-objective problem? 91 5.3 Overview of metaheuristics for the association rules mining problem 93 5.3.1 Generalities 93 5.3.2 Metaheuristics for categorical association rules 94 5.3.3 Evolutionary algorithms for quantitative association rules 99 5.3.4 Metaheuristics for fuzzy association rules 102 5.4 General table 105 5.5 Conclusion 107 Chapter 6 Metaheuristics and (Supervised) Classification 109 6.1 Task description and standard approaches 110 6.1.1 Problem description 110 6.1.2 K-nearest neighbor 110 6.1.3 Decision trees 111 6.1.4 Naive Bayes 112 6.1.5 Artificial neural networks 113 6.1.6 Support vector machines 114 6.2 Optimization model 114 6.2.1 A combinatorial problem 114 6.2.2 Quality measures 114 6.2.3 Methodology of performance evaluation in supervised classification 117 6.3 Metaheuristics to build standard classifiers 118 6.3.1 Optimization of K-NN 118 6.3.2 Decision tree 119 6.3.3 Optimization of ANN 122 6.3.4 Optimization of SVM 124 6.4 Metaheuristics for classification rules 126 6.4.1 Modeling 126 6.4.2 Objective function(s) 127 6.4.3 Operators 129 6.4.4 Algorithms 130 6.5 Conclusion 132 Chapter 7 On the Use of Metaheuristics for Feature Selection in Classification 135 7.1 Task description 136 7.1.1 Filter models 136 7.1.2 Wrapper models 137 7.1.3 Embedded models 137 7.2 Optimization model 138 7.2.1 A combinatorial optimization problem 138 7.2.2 Representation 139 7.2.3 Operators 140 7.2.4 Quality measures 140 7.2.5 Validation 143 7.3 Overview of methods 143 7.4 Conclusion 144 Chapter 8 Frameworks 147 8.1 Frameworks for designing metaheuristics 147 8.1.1 Easylocal++ 148 8.1.2 HeuristicLab 148 8.1.3 jMetal 149 8.1.4 Mallba 149 8.1.5 ParadisEO 150 8.1.6 ECJ 150 8.1.7 OpenBeagle 151 8.1.8 JCLEC 151 8.2 Framework for data mining 151 8.2.1 Orange 152 8.2.2 R and Rattle GUI 153 8.3 Framework for data mining with metaheuristics 153 8.3.1 RapidMiner 154 8.3.2 Weka 154 8.3.3 Keel 155 8.3.4 MO-Mine 157 8.4 Conclusion 157 Conclusion 159 Bibliography 161 Index 187
