Ensembles Of Type 2 Fuzzy Neural Models And Their Optimization With Bio-inspired Algorithms For Time Series Prediction (springerbriefs In Applied Sciences And Technology)

"This book focuses on the fields of hybrid intelligent systems based on fuzzy systems, neural networks, bio-inspired algorithms and time series. This book describes the construction of ensembles of Interval Type-2 Fuzzy Neural Networks models and the optimization of their fuzzy integrators with bio-inspired algorithms for time series prediction. Interval type-2 and type-1 fuzzy systems are used to integrate the outputs of the Ensemble of Interval Type-2 Fuzzy Neural Network models. Genetic Algorithms and Particle Swarm Optimization are the Bio-Inspired algorithms used for the optimization of the fuzzy response integrators. The Mackey-Glass, Mexican Stock Exchange, Dow Jones and NASDAQ time series are used to test of performance of the proposed method. Prediction errors are evaluated by the following metrics: Mean Absolute Error, Mean Square Error, Root Mean Square Error, Mean Percentage Error and Mean Absolute Percentage Error. The proposed prediction model outperforms state of the art methods in predicting the particular time series considered in this work" -- Proporcionat per l'editor Preface 6 Contents 8 1 Introduction 10 References 11 2 State of the Art 13 2.1 Time Series 13 2.2 Interval Type-2 Fuzzy Neural Network 14 2.3 Ensemble Learning 17 2.4 Interval Type-2 Fuzzy Systems 17 2.5 Genetic Algorithms 18 2.6 Particle Swarm Optimization 20 References 21 3 Problem Statement and Development 24 3.1 Historical Data 25 3.1.1 Mackey-Glass Time Series 26 3.1.2 Mexican Stock Exchange 27 3.1.3 Dow Jones Time Series 27 3.1.4 NASDAQ Time Series 28 3.2 Ensembles of IT2FNN Architectures 28 3.2.1 IT2FNN-1 Model 29 3.2.2 IT2FNN-2 Model 32 3.2.3 IT2FNN-3 Model 33 3.3 Fuzzy Integrators 35 3.4 Optimization of the Fuzzy Integration with the Genetic Algorithm 37 3.5 Optimization of the Fuzzy Integrators with the Particle Swarm Optimization 39 References 40 4 Simulation Studies 42 4.1 Mackey-Glass Time Series 42 4.1.1 Ensemble of the IT2FNN Architectures for Mackey-Glass 42 4.1.1.1 IT2FNN-1 Model 44 4.1.1.2 IT2FNN-2 Model 44 4.1.1.3 IT2FNN-3 Model 44 4.1.2 Optimization of the Fuzzy Integrators with the Genetic Algorithm 52 4.1.3 Optimization of the Fuzzy Integrators with the Particle Swarm Optimization 61 4.2 Mexican Stock Exchange Time Series 75 4.2.1 Ensemble of IT2FNN Architectures for BMV Time Series 75 4.2.1.1 IT2FNN-1 Model 75 4.2.1.2 IT2FNN-2 Model 77 4.2.1.3 IT2FNN-3 Model 79 4.3 Dow Jones Time Series 80 4.3.1 Ensemble of IT2FNN Architectures for Dow Jones Time Series 80 4.3.1.1 IT2FNN-1 Model 81 4.3.1.2 IT2FNN-2 Model 83 4.3.1.3 IT2FNN-3 Model 83 4.4 NASDAQ Time Series 86 4.4.1 Ensemble of IT2FNN Architectures for NASDAQ Time Series 86 4.4.1.1 IT2FNN-1 Model 87 4.4.1.2 IT2FNN-2 Model 88 4.4.1.3 IT2FNN-3 Model 90 4.5 Statistical Comparison Results of the Optimization of the Fuzzy Integrators 90 5 Conclusion 94 Appendix 96 IT2FNN-1 (Source Code) 96 Index 103 Annotation This book focuses on the fields of hybrid intelligent systems based on fuzzy systems, neural networks, bio-inspired algorithms and time series. This book describes the construction of ensembles of Interval Type-2 Fuzzy Neural Networks models and the optimization of their fuzzy integrators with bio-inspired algorithms for time series prediction. Interval type-2 and type-1 fuzzy systems are used to integrate the outputs of the Ensemble of Interval Type-2 Fuzzy Neural Network models. Genetic Algorithms and Particle Swarm Optimization are the Bio-Inspired algorithms used for the optimization of the fuzzy response integrators. The Mackey-Glass, Mexican Stock Exchange, Dow Jones and NASDAQ time series are used to test of performance of the proposed method. Prediction errors are evaluated by the following metrics: Mean Absolute Error, Mean Square Error, Root Mean Square Error, Mean Percentage Error and Mean Absolute Percentage Error. The proposed prediction model outperforms state of the art methods in predicting the particular time series considered in this work Front Matter ....Pages i-viii Introduction (Jesus Soto, Patricia Melin, Oscar Castillo)....Pages 1-3 State of the Art (Jesus Soto, Patricia Melin, Oscar Castillo)....Pages 5-15 Problem Statement and Development (Jesus Soto, Patricia Melin, Oscar Castillo)....Pages 17-34 Simulation Studies (Jesus Soto, Patricia Melin, Oscar Castillo)....Pages 35-86 Conclusion (Jesus Soto, Patricia Melin, Oscar Castillo)....Pages 87-88 Back Matter ....Pages 89-97