The accuracy of a procedural approach to specifying feedforward neural networks for forecasting

Liao, Kua-Ping; Fildes, Robert

doi:10.1016/j.cor.2004.02.006

Cited by 40 publications

(24 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, Adya and Collopy (1998) found eleven studies that met the criteria for a comparative evaluation, and in 8 of these (73%), neural nets were more accurate. There were no estimates of the error reductions versus alternative methods although Liao and Fildes (2005), in a test involving 261 series, 18 horizons, and 5 forecast origins, found impressive gains in accuracy for neural nets with an error reduction of 56% compared to damped trends. Chatfield (personal correspondence) suspects that there is a 'file-drawer problem,' saying that he knew of some studies that failed to show gains and were not submitted for publication, and there is a well-known bias by reviewers against papers with null results.…”

Section: Neural Netsmentioning

confidence: 99%

See 1 more Smart Citation

Findings from evidence-based forecasting: Methods for reducing forecast error

Armstrong

2006

International Journal of Forecasting

151

View full text Add to dashboard Cite

Empirical comparisons of reasonable approaches provide evidence on the best forecasting procedures to use under given conditions. Based on this evidence, I summarize the progress made over the past quarter century with respect to methods for reducing forecasting error. Seven well-established methods have been shown to improve accuracy: combining forecasts and Delphi help for all types of data; causal modeling, judgmental bootstrapping and structured judgment help with cross-sectional data; and causal models and trend-damping help with time-series data. Promising methods for cross-sectional data include damped causality, simulated interaction, structured analogies, and judgmental decomposition; for time-series data, they include segmentation, rule-based forecasting, damped seasonality, decomposition by causal forces, damped trend with analogous data, and damped seasonality. The testing of multiple hypotheses has also revealed methods where gains are limited: these include data mining, neural nets, and Box-Jenkins methods. Multiple hypotheses tests should be conducted on widely used but relatively untested methods such as prediction markets, conjoint analysis, diffusion models, and game theory. KeywordsBox-Jenkins, causal forces, causal models, combining forecasts, complex series, conjoint analysis, contrary series, damped seasonality, damped trend, data mining, Delphi, diffusion, game theory, judgmental decomposition, multiple hypotheses, neural nets, prediction markets, rule-based forecasting, segmentation, simulated interaction, structured analogies AbstractEmpirical comparisons of reasonable approaches provide evidence on the best forecasting

show abstract

Section: Neural Netsmentioning

confidence: 99%

“…On the other hand, the impressive findings from Liao and Fildes (2005) deserve further attention in an effort to discover the conditions under which neural nets are useful. For the latest on neural nets, see the special interest group at forecastingprinciples.com.…”

Section: Neural Netsmentioning

confidence: 99%

Findings from evidence-based forecasting: Methods for reducing forecast error

Armstrong

2006

International Journal of Forecasting

151

View full text Add to dashboard Cite

show abstract

“…'Fuzzy set' approaches have been included here.) As we discuss in Section 1.3, these have been primarily applied to crosssectional classification problems such as consumer credit risk (see the discussion in the following section); there have been only a limited number of applications to time series with conflicting results (see eg, Makridakis and Hibon (2000); Liao and Fildes (2005), the former negative, the latter positive).…”

Section: Extrapolative Methodsmentioning

confidence: 99%

“…The majority of studies involve the development of computer intensive methods able to model and forecast a single time series. However, there are only a small number of studies where computing power has been used to tackle large problems, either on cross-sectional data (with applications in CRM, see previous section) or in time series (eg, Liao and Fildes (2005), who examined various specifications of ANN compared to benchmarks, and Terasvirta et al (2005), who compared ANN with non-linear statistical models). But such papers present offline analyses that have yet to find their way into practice.…”

Section: The Role Of Computer and Is Developments 231 It Is And Fmentioning

confidence: 99%

Forecasting and operational research: a review

Fildes

Νικολόπουλος

Crone

et al. 2008

Journal of the Operational Research Society

225

View full text Add to dashboard Cite

From its foundation, operational research (OR) has made many substantial contributions to practical forecasting in organizations. Equally, researchers in other disciplines have influenced forecasting practice. Since the last survey articles in JORS, forecasting has developed as a discipline with its own journals. While the effect of this increased specialization has been a narrowing of the scope of OR's interest in forecasting, research from an OR perspective remains vigorous. OR has been more receptive than other disciplines to the specialist research published in the forecasting journals, capitalizing on some of their key findings. In this paper, we identify the particular topics of OR interest over the past 25 years. After a brief summary of the current research in forecasting methods, we examine those topic areas that have grabbed the attention of OR researchers: computationally intensive methods and applications in operations and marketing. Applications in operations have proved particularly important, including the management of inventories and the effects of sharing forecast information across the supply chain. The second area of application is marketing, including customer relationship management using data mining and computer-intensive methods. The paper concludes by arguing that the unique contribution that OR can continue to make to forecasting is through developing models that link the effectiveness of new forecasting methods to the organizational context in which the models will be applied. The benefits of examining the system rather than its separate components are likely to be substantial.

show abstract

“…Recently, both CI-methods of Multilayer Perceptrons (MLP) and Support Vector Regression (SVR) have shown promising performance in various scientific forecasting domains [1][2], offering non parametric, data-driven and self-adaptive approaches that learn linear or nonlinear functional relationships directly from data [3][4]. In order to objectively prove the efficacy of CI-algorithms in forecasting, outside of a controlled research experiment in which the test data is known to the researcher, their accuracy must be evaluated in a series of true ex ante comparisons against established statistical forecasting methods on empirical datasets [3,5]. The 20 10 Neural Network Grand Challenge (NNGC) provides a further opportunity to establish the forecasting accuracy of CI on six datasets of 11 empirical time series of transportation data.…”

Section: Introductionmentioning

confidence: 99%