A Rule-based Forecasting System Integrating combining and Single Forecast for Decision Making

Chun‐Ling Lin, Chen‐Chun Lin, Joseph Z. Shyu, Chin‐Teng LinInternational Journal of Computer Applications 2012 · DOI

Abstract

Many studies demonstrated that combining forecasts produces consistent but modest gains in accuracy. However, little researches define well the conditions under neither which combining is most effective nor how methods should be combined in each situation. In this paper, a rule-based forecasting system industry is proposed in order to compare forecast performance between combining forecasts and single forecasts, then define these conditions and to specify more effective combinations, finally suggest the best methods. Two comparative case studies for the telecommunications and TFT-LCD industry are proposed to examine the performance of the proposed system. Results from this study indicate that combining forecasts outperform single forecasts only when data set is data have various nonlinear characteristics. In this research, empirical evidence shows that rules based on causal forces improved the selection of forecasting methods, the structuring of time series, and the assessment of prediction intervals.

Introduction

In the field of time-series forecasting, selecting the optimal model remains a significant challenge. No single approach—whether a classical statistical method like ARIMA or a modern machine learning technique like Neural Networks—is universally superior for all datasets. While combining multiple forecasts generally improves accuracy, there is a lack of clear guidance on when combination is most effective and which specific methods should be paired. This paper introduces a Rule-Based Forecasting (RBF) system designed to bridge this gap by utilizing predefined rules to select the most effective forecasting strategy based on specific data characteristics.

System Architecture and Methodology

The proposed RBF system integrates four distinct forecasting modules managed through a centralized control engine. The architecture is designed to incorporate domain knowledge and expert rules into the automated selection process.

Architecture of the Rule-based Forecasting System

The system consists of several key subsystems: a Database Management subsystem for handling demand and revenue data, a Model Management subsystem containing various forecasting models, a Knowledge Management subsystem for storing selection rules, and a Dialogue Management subsystem providing a user interface for decision-makers.

The forecasting process follows a structured logic to identify the optimal model for a given dataset, as illustrated in the flowchart below.

Flowchart of the forecasting selection and combination process

The forecasting modules include:

  • Single Models: Linear methods (ARIMA, Exponential Smoothing) and nonlinear methods (BPNN, ANFIS, SVR).
  • Hybrid Linear-Nonlinear Models: These models apply a linear method first and then use a nonlinear method to analyze the residuals (e.g., ARIMA_BPNN).
  • Linear Combining: A module that merges the top three performing models using linear regression.
  • Nonlinear Combining: Employs ANFIS or BPNN to combine outputs from the top three models.

Case Studies and Results

The system was validated using two industrial datasets: 3G telecommunications demand in Taiwan and global TFT-LCD industry revenue (segmented by Asia Pacific, China, and North America). Performance was measured using Mean Square Error (MSE).

Key Findings

  • Nonlinearity as a Determinant: The study found that combining forecasts is not a universal solution. It consistently outperforms single models only when the data exhibits high nonlinearity and significant fluctuations.
  • Efficiency of Single Models: In the telecommunications case study, which featured relatively linear data, a single Support Vector Regression (SVR) model achieved the best performance ($MSE = 1.98 \times 10^{11}$), outperforming more complex combinations. This suggests that "over-engineering" forecasts can be counterproductive for simpler data trends.
  • Regional Variability: In the TFT-LCD study, the optimal model varied by region. For Asia Pacific, a hybrid ES_SVR model was superior ($MSE = 7.17 \times 10^9$), while China's more volatile data favored an ANFIS nonlinear combination ($MSE = 7.89 \times 10^{11}$). For North America, the single SVR model again provided the most accurate forecast ($MSE = 5.44 \times 10^{10}$).

Conclusion

The RBF system demonstrates that a rule-based approach can effectively translate forecasting expertise into automated selection criteria. Empirical results suggest that while combined models are powerful for complex, nonlinear data, single robust models like SVR may be more efficient for linear trends. Future work will focus on incorporating broader external causal factors to further refine the selection rules across diverse industries.

Summary updated

Abstract

Many studies demonstrated that combining forecasts produces consistent but modest gains in accuracy. However, little researches define well the conditions under neither which combining is most effective nor how methods should be combined in each situation. In this paper, a rule-based forecasting system industry is proposed in order to compare forecast performance between combining forecasts and single forecasts, then define these conditions and to specify more effective combinations, finally suggest the best methods. Two comparative case studies for the telecommunications and TFT-LCD industry are proposed to examine the performance of the proposed system. Results from this study indicate that combining forecasts outperform single forecasts only when data set is data have various nonlinear characteristics. In this research, empirical evidence shows that rules based on causal forces improved the selection of forecasting methods, the structuring of time series, and the assessment of prediction intervals.
BibTeX
@article{lin2012rule,
  author = {Chun‐Ling Lin and Chen‐Chun Lin and Joseph Z. Shyu and Chin‐Teng Lin},
  title = {{A Rule-based Forecasting System Integrating combining and Single Forecast for Decision Making}},
  journal = {International Journal of Computer Applications},
  year = {2012},
  doi = {10.5120/4885-7195},
}
Figure