#### DMCA

## Neural Comput & Applic DOI 10.1007/s00521-011-0741-0 (2011)

### Citations

1248 | Approximations by superpositions of a sigmoidal function
- Cybenko
- 1989
(Show Context)
Citation Context ...eptron (MLP) as computational model due to its approximation capability and inside this group, Full Connected MLP with only a hidden layer and Backpropagation (BP) as learning algorithm, according to =-=[31]-=-. As it was mentioned in Sect. 1, the design of the ANN will be done by setting the parameters of the ANN. In the case of MLP with only one hidden layer and BP, these parameters are: number of inputs ... |

842 |
Evolutionary computation: towards a new philosophy of machine intelligence,
- Fogel
- 1996
(Show Context)
Citation Context ...NN with 3 input nodes (k = 3) can be seen at Fig. 1. 3.2 ADANN The problem of designing ANN could be seen as a search problem into the space of all possible ANN. Then, that search can be done by a GA =-=[30]-=- using exploitation and exploration. Therefore, there are three crucial issues: (1) about the solution’s space, which information of the net should be included into the chromosome; (2) how this inform... |

436 |
Differential evolution-a simple and efficient heuristic for global optimization over continuous spaces
- Storn, Price
- 1997
(Show Context)
Citation Context ...d to forecast the next one, t ? 1(1to N ahead forecast). 3.3 Differential evolution algorithm Differential evolution algorithm (DE) is a stochastic nonlinear optimization algorithm by Storn and Price =-=[35]-=- and belongs to the class of evolution strategy optimizers. DE looks for the global minimum of a multidimensional, multimodal function trying to obtain a good probability. 123Neural Comput & Applic D... |

356 | Designing neural networks using genetic algorithms with graph generation system. - Kitano - 1990 |

248 |
Genetic algorithms in search optimization and machine learning.
- DE
- 1989
(Show Context)
Citation Context ...nly training data and a few datasets) have shown that the MSE fitness leads to better forecasts when compared with mean absolute error (MAE). To choose the genetic algorithm parameters, Goldberg’s GA =-=[33]-=- and [34] were taken into account, apart of the experience obtained during the preliminary experiments. The GA parameters were set to population size, 50; maximum number of Fig. 3 ANN design by GA sch... |

229 |
Forecasting with artificial neural networks: The state of the art.
- Zhang, Patuwo, et al.
- 1998
(Show Context)
Citation Context ...ct. 5. 2 Related work Several authors have addressed forecasting tasks using ANN [4, 7, 12, 13]. This reveals the full consideration of ANN (as a data driven learning machine) into forecasting theory =-=[14, 15]-=-. Often, the process of designing the correct ANN model for TSF is based on trial and error heuristics. If manual design of ANN is carried out, several topologies (i.e. different number of inputs node... |

202 | A new evolutionary system for evolving artificial neural networks. - Yao, Liu - 1997 |

202 | A review of evolutionary artificial neural networks.
- Yao
- 1993
(Show Context)
Citation Context ... so that a constructive method gives rise to an artificial neural network topology (or architecture). Abraham [21] shows an automatic framework for optimization ANN in an adaptive way, and Yao et al. =-=[22]-=- try to spell out the future trends of the field. About the use of EANN in time series forecasting, many works have been carried out [23–25]. In [14], Patuwo and Hu present a ‘‘state of the art’’ of A... |

152 |
Accessed on:
- htm
- 2014
(Show Context)
Citation Context ...ation 60 and 70, DE and EDA keeps on decreasing their values after generation 100. 4.4 ADANN versus Forecast pro Ò As a baseline comparison, we choose the popular forecasting tool Forecast Pro Ò (FP) =-=[41]-=-. In particular, we fed the tool with the in-samples of five datasets (i.e. Passengers, Temperature, Dow-Jones, Quebec and Mackey– Glass) and executed the full automatic feature of the tool to obtain ... |

143 |
Dynamic node creation in back propagation networks.
- Ash
- 1989
(Show Context)
Citation Context ... automatic ANN design, where EC plays an important role, in what is known as evolutionary ANN (EANN). Several works have proposed EANN, such as [16–20]. Some of them use direct encoding schemes (DES) =-=[16, 17]-=-, and the others use Indirect Encoding Scheme (IES) [18–20]. For DES, the chromosome contains information about parameters of the topology, architecture, learning parameters, etc. of the artificial ne... |

85 |
Long-Range Forecasting.
- Armstrong
- 1985
(Show Context)
Citation Context ...y popular metric within TSF. SMAPE has the advantage of being scale independent, thus can be more easily used to compare methods across different series. Although the SMAPE was originally proposed in =-=[39]-=-, (4) adopts the variant used in [40] since it does not lead to negative values (ranging from 0 to 200%). SMAPE is also used in NN3, NN5 and NNGC1 [38] forecasting competitions as evaluation error. Fo... |

61 |
Neural Networks: A Comprehensive Foundation,
- Simon
- 1999
(Show Context)
Citation Context ...tor machines (SVM) [3], artificial neural networks (ANN) [4], fuzzy techniques [5] or hybrid systems combining any of the previous ones with evolutionary search [6, 7]. In this paper, we focus on ANN =-=[11]-=- which are flexible models that do not require a priori knowledge, are capable of nonlinear modeling and also often robust to noisy data. These properties of ANN make them a natural solution for TSF. ... |

61 | Genetic Synthesis of Boolean Neural Networks with a Cell Rewriting Developmental Process. - Gruau - 1992 |

46 | Meta learning evolutionary artificial neural networks,”
- Abraham
- 2004
(Show Context)
Citation Context ...f the artificial neural network. In IES the chromosome contains the necessary information, so that a constructive method gives rise to an artificial neural network topology (or architecture). Abraham =-=[21]-=- shows an automatic framework for optimization ANN in an adaptive way, and Yao et al. [22] try to spell out the future trends of the field. About the use of EANN in time series forecasting, many works... |

15 |
A fusion model of HMM, ANN and GA for stock market forecasting,”
- Hassan, Nath, et al.
- 2007
(Show Context)
Citation Context ...odeling and also often robust to noisy data. These properties of ANN make them a natural solution for TSF. In effect, ANN have been applied in real-world forecasting tasks, such as market predictions =-=[8]-=-, meteorological [9] and network traffic forecasting [10]. A crucial issue is the design of an appropriate ANN model for a particular time series [4]. It will be necessary to 123Neural Comput & Appli... |

10 | Evolving Time Series Forecasting ARMA Models.
- Cortez, Rocha, et al.
- 2004
(Show Context)
Citation Context ...ct. 5. 2 Related work Several authors have addressed forecasting tasks using ANN [4, 7, 12, 13]. This reveals the full consideration of ANN (as a data driven learning machine) into forecasting theory =-=[14, 15]-=-. Often, the process of designing the correct ANN model for TSF is based on trial and error heuristics. If manual design of ANN is carried out, several topologies (i.e. different number of inputs node... |

10 |
2001 Estimation Of Distribution Algorithms: A New Tool For Evolutionary Computation
- Larranaga, Lozano
(Show Context)
Citation Context ... GA, which is responsible for carrying out the global search into the hybrid system, by DE as it can be seen in [44]. 3.4 Estimation distribution algorithm Estimation of distribution algorithms (EDA) =-=[42]-=- is an outgrowth of GA. In GA, a population of individual solutions of a problem is kept as part of the search for a better solution. This population is typically represented as an array of objects. D... |

8 | The application of antigenic search techniques to time series forecasting
- Nunn, White
(Show Context)
Citation Context ...d, such as HoltWinters (in the sixties) or the ARIMA methodology [1] (in the seventies). More recently, several computational intelligence (CI) methods have been applied to TSF such as immune systems =-=[2]-=-, support vector machines (SVM) [3], artificial neural networks (ANN) [4], fuzzy techniques [5] or hybrid systems combining any of the previous ones with evolutionary search [6, 7]. In this paper, we ... |

8 | Choosing Leaders for Multi-objective PSO Algorithms Using Differential Evolution
- Wickramasinghe, Li
(Show Context)
Citation Context ...mensional, multimodal function trying to obtain a good probability. 123Neural Comput & Applic DE community has been growing since the mid 1990s, and today more researchers are working on and with DE =-=[29, 36]-=-. The main idea of DE is a scheme for generating trial parameter vectors. DE differs from other evolutionary algorithms (EA) in their mechanism of generating offspring. In GA, an individual plays the ... |

6 | Time series forecasting by evolutionary neural networks - CORTEZ, ROCHA, et al. - 2005 |

5 |
Forecasting: methods and applications, 2nd edn
- Makridakis, Wheelwright, et al.
- 1983
(Show Context)
Citation Context ... become increasingly used in areas such as agriculture, finance, management, production or sales. Several TSF methods have been proposed, such as HoltWinters (in the sixties) or the ARIMA methodology =-=[1]-=- (in the seventies). More recently, several computational intelligence (CI) methods have been applied to TSF such as immune systems [2], support vector machines (SVM) [3], artificial neural networks (... |

5 |
Stepwise selection of artificial neural networks models for time series prediction
- SF
(Show Context)
Citation Context ...es. In Sect. 4, experimental setup and results are shown. And finally, conclusions and future works are described in Sect. 5. 2 Related work Several authors have addressed forecasting tasks using ANN =-=[4, 7, 12, 13]-=-. This reveals the full consideration of ANN (as a data driven learning machine) into forecasting theory [14, 15]. Often, the process of designing the correct ANN model for TSF is based on trial and e... |

2 |
Sensitivity analysis for time lag selection to forecast seasonal time series using neural networks and support vector machines
- Cortez
- 2010
(Show Context)
Citation Context ...ies) or the ARIMA methodology [1] (in the seventies). More recently, several computational intelligence (CI) methods have been applied to TSF such as immune systems [2], support vector machines (SVM) =-=[3]-=-, artificial neural networks (ANN) [4], fuzzy techniques [5] or hybrid systems combining any of the previous ones with evolutionary search [6, 7]. In this paper, we focus on ANN [11] which are flexibl... |

2 |
Kourentzes N (2010) Feature selection for time series prediction a combined filter and wrapper approach for neural networks. Neurocomputing 73:1923–1936
- Crone
(Show Context)
Citation Context ...the seventies). More recently, several computational intelligence (CI) methods have been applied to TSF such as immune systems [2], support vector machines (SVM) [3], artificial neural networks (ANN) =-=[4]-=-, fuzzy techniques [5] or hybrid systems combining any of the previous ones with evolutionary search [6, 7]. In this paper, we focus on ANN [11] which are flexible models that do not require a priori ... |

2 |
těpnička M, Dvorˇák A, Pavliska V, Vavrˇíčková L (2011) A linguistic approach to time series modeling with help of the F-transform. Fuzzy Sets Syst 180(1):164–184
- unknown authors
(Show Context)
Citation Context ...ecently, several computational intelligence (CI) methods have been applied to TSF such as immune systems [2], support vector machines (SVM) [3], artificial neural networks (ANN) [4], fuzzy techniques =-=[5]-=- or hybrid systems combining any of the previous ones with evolutionary search [6, 7]. In this paper, we focus on ANN [11] which are flexible models that do not require a priori knowledge, are capable... |

2 |
Song Q (2002) Dens: Dynamic evolving neural-fuzzy inference system and its application for time-series prediction
- Kasabov
(Show Context)
Citation Context ...such as immune systems [2], support vector machines (SVM) [3], artificial neural networks (ANN) [4], fuzzy techniques [5] or hybrid systems combining any of the previous ones with evolutionary search =-=[6, 7]-=-. In this paper, we focus on ANN [11] which are flexible models that do not require a priori knowledge, are capable of nonlinear modeling and also often robust to noisy data. These properties of ANN m... |

2 |
Sanchis A (2008) Adann: automatic design of artificial neural networks
- Peralta, Gutierrez
(Show Context)
Citation Context ...such as immune systems [2], support vector machines (SVM) [3], artificial neural networks (ANN) [4], fuzzy techniques [5] or hybrid systems combining any of the previous ones with evolutionary search =-=[6, 7]-=-. In this paper, we focus on ANN [11] which are flexible models that do not require a priori knowledge, are capable of nonlinear modeling and also often robust to noisy data. These properties of ANN m... |

2 |
Abraham A (2007) An ensemble of neural networks for weather forecasting. Neural Comput Appl 13(2):112–122
- Maqsood, MR
(Show Context)
Citation Context ...en robust to noisy data. These properties of ANN make them a natural solution for TSF. In effect, ANN have been applied in real-world forecasting tasks, such as market predictions [8], meteorological =-=[9]-=- and network traffic forecasting [10]. A crucial issue is the design of an appropriate ANN model for a particular time series [4]. It will be necessary to 123Neural Comput & Applic set the type of AN... |

2 |
Fessant F, Collobert D
- Bengio
- 1995
(Show Context)
Citation Context ...rties of ANN make them a natural solution for TSF. In effect, ANN have been applied in real-world forecasting tasks, such as market predictions [8], meteorological [9] and network traffic forecasting =-=[10]-=-. A crucial issue is the design of an appropriate ANN model for a particular time series [4]. It will be necessary to 123Neural Comput & Applic set the type of ANN that will solve the forecasting tas... |

2 |
Adeodato PJ (2008) A new intelligent system methodology for time series forecasting with artificial neural networks
- TA, GC
(Show Context)
Citation Context ...es. In Sect. 4, experimental setup and results are shown. And finally, conclusions and future works are described in Sect. 5. 2 Related work Several authors have addressed forecasting tasks using ANN =-=[4, 7, 12, 13]-=-. This reveals the full consideration of ANN (as a data driven learning machine) into forecasting theory [14, 15]. Often, the process of designing the correct ANN model for TSF is based on trial and e... |

2 |
Evolving neural network. Biol Cybern 63:487–493
- DB, LJ, et al.
- 1990
(Show Context)
Citation Context ... automatic ANN design, where EC plays an important role, in what is known as evolutionary ANN (EANN). Several works have proposed EANN, such as [16–20]. Some of them use direct encoding schemes (DES) =-=[16, 17]-=-, and the others use Indirect Encoding Scheme (IES) [18–20]. For DES, the chromosome contains information about parameters of the topology, architecture, learning parameters, etc. of the artificial ne... |

2 | Hiltunena T, Karppinenb A, Ruuskanena J Kolehmainena M (2004) Evolving the neural network model for forecasting air pollution time series. Eng Appl Artif Intell 17(2):159–167. ISSN - Niska |

2 | Chang F-J (2009) Evolutionary artificial neural networks for hydrological systems forecasting - Chena |

2 |
Dong J (2005) Time-series prediction using a local linear wavelet neural network. Neurocomputing 69(4–6): 449–465
- Chen, Yang
(Show Context)
Citation Context ...ut [23–25]. In [14], Patuwo and Hu present a ‘‘state of the art’’ of ANN into forecasting task, and in [13], Crone proposes a stepwise selection of ANN models for time series forecasting. Chen et al. =-=[26]-=- propose local linear wavelet neural network to forecast time series. Rivas et al. [27] use evolving RBF neural networks for time series forecasting. Few studies have been done using ANN and DE to gen... |

2 |
JG (2003) Evolving RBF neural networks for time-series forecasting with EvRBF. Inform Sci 165(3–4):207–220
- VM, JJ, et al.
(Show Context)
Citation Context ...ng task, and in [13], Crone proposes a stepwise selection of ANN models for time series forecasting. Chen et al. [26] propose local linear wavelet neural network to forecast time series. Rivas et al. =-=[27]-=- use evolving RBF neural networks for time series forecasting. Few studies have been done using ANN and DE to generate a hybrid system to forecast time series [28, 29]. Since EDA is a more recent tech... |

2 |
Vasconselos GC, Ferreria AE (2007) Hybrid differential evolutionary system for financial time series forecasting
- RA
(Show Context)
Citation Context ...o forecast time series. Rivas et al. [27] use evolving RBF neural networks for time series forecasting. Few studies have been done using ANN and DE to generate a hybrid system to forecast time series =-=[28, 29]-=-. Since EDA is a more recent technique, its use for EANN in TSF is scarce and within our knowledge, has only appeared very recently. Therefore, the main motivation of this paper is to obtain ANN model... |

2 |
Neural networks training based on differential evolution algorithm compared with other architectures for weather forecasting 34. Int J Comput Sci Netw Secur 9(3):92–99
- HM
- 2009
(Show Context)
Citation Context ...o forecast time series. Rivas et al. [27] use evolving RBF neural networks for time series forecasting. Few studies have been done using ANN and DE to generate a hybrid system to forecast time series =-=[28, 29]-=-. Since EDA is a more recent technique, its use for EANN in TSF is scarce and within our knowledge, has only appeared very recently. Therefore, the main motivation of this paper is to obtain ANN model... |

2 |
Sagarna R, Larrañaga P (2002) Adjusting weights in artificial neural networks using evolutionary algorithms. In: Larrañaga P, Lozano JA (eds) Estimation of distribution algorithms. A new tool for evolutionary computation
- Cotta, Alba
(Show Context)
Citation Context ...ecause estimation distribution algorithm is a recent technique (came into use just a few years ago), it has been carried out few hybrid studies (i.e. ANN ? EDA) applied only to classification domains =-=[37]-=-. In this paper, it is proposed a new hybrid method using advantages of EDA to design ANN to forecast time series. Besides, it is a totally automatic method. There are different kinds of EDA, but for ... |

2 |
Atiya A (2009) A new Bayesian formulation for Holt’s exponential smoothing
- Andrawis
(Show Context)
Citation Context ...s the advantage of being scale independent, thus can be more easily used to compare methods across different series. Although the SMAPE was originally proposed in [39], (4) adopts the variant used in =-=[40]-=- since it does not lead to negative values (ranging from 0 to 200%). SMAPE is also used in NN3, NN5 and NNGC1 [38] forecasting competitions as evaluation error. For the forecasting comparison, we opte... |

2 |
Using genetic search to exploit the emergent behavior of neural networks
- J, RA, et al.
- 1991
(Show Context)
Citation Context ...are reached. A schema of the whole search process can be seen at Fig. 3. The fitness value for each individual is the minimum validation error during the learning process (ANN training), according to =-=[43]-=-. Regarding the use of MSE in the fitness function, the rationale is to reduce extreme errors that may highly affect multistep ahead forecasts. Also, preliminary experiments (using only training data ... |

2 |
Sanchis A (2010) Time series forecasting by evolving artificial neural networks using genetic algorithms and estimation of distribution algorithms
- Peralta, Gutierrez
(Show Context)
Citation Context ...ried out between X i and U i. To apply DE to our system, it was necessary to replace the GA, which is responsible for carrying out the global search into the hybrid system, by DE as it can be seen in =-=[44]-=-. 3.4 Estimation distribution algorithm Estimation of distribution algorithms (EDA) [42] is an outgrowth of GA. In GA, a population of individual solutions of a problem is kept as part of the search f... |

2 |
Time series data library, http://robjhyndman. com/TSDL/. Accessed on
- RJ
- 2011
(Show Context)
Citation Context ...N5 (2008) time series competition [38]. To compare the proposed TSF methods, we selected 5 benchmark time series. Four of them are series from the well-known Hyndman’s Time Series Data Library (TSDL) =-=[46]-=-. These time series are named Passengers, Temperature, Dow-Jones and Quebec. Passengers time series represents the number of passengers of an international airline in thousands, measured monthly from ... |