MLP ensembles improve long term prediction accuracy over single networks

Adeodato, Paulo J. L.; Arnaud, Adrian L.; Vasconcelos, Germano C.; Cunha, Rodrigo C.L.V.; Monteiro, Domingos S. M. P.

doi:10.1016/j.ijforecast.2009.05.029

Cited by 49 publications

(22 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The ordering phase usually requires setting high learning rate, The simple ANN mode was referred to as "ANN I" and the modified version as "ANN II", as depicted earlier in Figure 1. For both modes, input data were divided randomly into a training, validation and test subsets in the following fashion 70%, 15% and 15% respectively, for weight learning, over-fitting prevention and performance validation [1]. The network was trained using the Levenberg Marquardt backpropagation algorithm, which is known as one of the most effecient ANN training algorithem [17] [2].…”

Section: Artificial Neural Network Techniquesmentioning

confidence: 99%

Unified knowledge based economy hybrid forecasting

Shami

Lotfi

Coleman

et al. 2015

Technological Forecasting and Social Change

View full text Add to dashboard Cite

Section: Artificial Neural Network Techniquesmentioning

confidence: 99%

Unified knowledge based economy hybrid forecasting

Shami

Lotfi

Coleman

et al. 2015

Technological Forecasting and Social Change

View full text Add to dashboard Cite

“…In ensemble construction, once the characteristics of the models are established, variety is usually introduced by modifying initial random weights ( [20]), or by randomising training samples ( [7], [21]). The randomisation of the feature space ( [9]) could count both as a strategy for model variation and as a strategy for model specification.…”

Section: Generatementioning

confidence: 99%

“…The methods that have been proposed in the literature (focusing on forecasting rather than on classification) include: a gating network ( [22]), a simple or weighted average ( [8], [12], [23], [24]), a nonlinear average through another NN ( [23]), a feed-forward NN ( [25], [26]), a Radial Basis Function ( [11]), the median of forecasts ( [10], [20]) and the mode ( [19]). It can be seen that the complexity and effectiveness of the ensembling approaches are only partially related to the combining procedure at the end of the process, as there are other steps involved.…”

Section: Generatementioning

confidence: 99%

Structural Combination of Neural Network Models

Juan¹,

Menezes

2016

2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW)

View full text Add to dashboard Cite

This is the accepted version of the paper.This version of the publication may differ from the final published version. ). This paper explores the incorporation of internal structure parameters of feed-forward neural network (NN) models as an approach to combine their forecasts via ensembles. First, the generated NN models that could be part of the ensembles are subject to a clustering algorithm that uses the structure parameters and, from each of the clusters obtained, a small set of models is selected and their forecasts are combined in a two-stage procedure. Secondly, in an alternative and simpler implementation, a subset of the generated NN models is selected by using several reference points in the model structure parameter space. The choice of the reference points is optimised through a genetic algorithm and the models selected are averaged. Hourly electricity demand time series is used to assess multi-step ahead forecasting performance for up to a 12 hours horizon. Results are compared against several statistical benchmarks, the average of the individual forecasts and the best models in the ensembles. Results show that the clusterbased (CB) structural combinations do better than the genetic algorithm (GA) structural combinations in outperforming the average forecast, which is the traditional point forecast from an ensemble. Permanent repository link

show abstract

“…In this work, the ensemble consisting of 31 Logistic Regression models has reduced the system's variance and their median was taken as the response for each test example. This median approach had been adopted by the authors' teams since 2007 in PAKDD Data Mining Competition [12] and in NN3 Time Series Forecasting Competition [13]. As already stated, the modeling technique chosen was Logistic Regression due to its explicit coefficients and for not having the need of a validation set.…”

Section: Logistic Regression and Model Ensemblementioning

confidence: 99%

Domain Driven Data Mining for Unavailability Estimation of Electrical Power Grids

Adeodato

Braga

Arnaud³

et al. 2010

Trends in Applied Intelligent Systems

Self Cite

View full text Add to dashboard Cite

Abstract. In Brazil, power generating, transmitting and distributing companies operating in the regulated market are paid for their equipment availability. In case of system unavailability, the companies are financially penalized, more severely, on unplanned interruptions. This work presents a domain driven data mining approach for estimating the risk of systems' unavailability based on their component equipments historical data, within one of the biggest Brazilian electric sector companies. Traditional statistical estimators are combined with the concepts of Recency, Frequency and Impact (RFI) for producing variables containing behavioral information finely tuned to the application domain. The unavailability costs are embedded in the problem modeling strategy. Logistic regression models bagged via their median score achieved Max_KS=0.341 and AUC_ROC=0.699 on the out-of-time data sample. This performance is much higher than the previous approaches attempted within the company. The system has been put in operation and will be monitored for the performance reassessment and maintenance re-planning.

show abstract

MLP ensembles improve long term prediction accuracy over single networks

Cited by 49 publications

References 11 publications

Unified knowledge based economy hybrid forecasting

Unified knowledge based economy hybrid forecasting

Structural Combination of Neural Network Models

Domain Driven Data Mining for Unavailability Estimation of Electrical Power Grids

Contact Info

Product

Resources

About