ANFIS Construction With Sparse Data via Group Rule Interpolation

Jing Yang, Changjing Shang, Ying Li, Fangyi Li, Qiang Shen

Abstract

A major assumption for constructing an effective adaptive-network-based fuzzy inference system (ANFIS) is that sufficient training data are available. However, in many real-world applications, this assumption may not hold, thereby requiring alternative approaches. In light of this observation, this article focuses on automated construction of ANFISs in an effort to enhance the potential of the Takagi-Sugeno fuzzy regression models for situations where only limited training data are available. In particular, the proposed approach works by interpolating a group of fuzzy rules in a certain given domain with the assistance of existing ANFISs in its neighboring domains. The construction process involves a number of computational mechanisms, including a rule dictionary which is created by extracting the rules from the existing ANFISs; a group of rules which are interpolated by exploiting the local linear embedding algorithm to build an intermediate ANFIS; and a fine-tuning method which refines the resulting intermediate ANFIS. The experimental evaluation on both synthetic and real-world datasets is reported, demonstrating that when facing the data shortage situations, the proposed approach helps significantly improve the performance of the original ANFIS modeling mechanism.

Introduction

Adaptive-Network-based Fuzzy Inference Systems (ANFIS) represent a powerful class of models that combine the transparent, human-like reasoning of fuzzy logic with the learning capabilities of neural networks. Specifically, these systems utilize Takagi-Sugeno (TSK) fuzzy rules, which are highly effective at approximating complex, nonlinear functions. However, a significant bottleneck in deploying ANFIS is its "data hunger." To build a robust model, a large and dense dataset is typically required.

In many practical applications—such as predicting rare medical events, analyzing niche engineering failures, or processing specific image patches—data is often sparse. When training data is thin, standard ANFIS construction often results in "holes" in the rule base, where the system lacks the necessary logic to respond to unseen inputs. While Fuzzy Rule Interpolation (FRI) techniques exist, they generally address single missing rules. This research proposes a method to construct an entire functional system simultaneously by leveraging knowledge from neighboring source domains.

Methodology: The Three-Stage Framework

The proposed approach moves beyond simple rule-filling to a comprehensive system construction framework consisting of three primary stages.

The three-stage framework for ANFIS construction: Rule Dictionary Generation, Intermediate ANFIS Construction, and Fine-Tuning.

1. Rule Dictionary Generation

Rather than initializing from scratch, the system extracts fuzzy rules from existing ANFIS models trained on related "source domains"—neighboring areas where data is plentiful. These rules are decomposed into their antecedent parts (the if conditions) and consequent parts (the then results), forming a searchable "Rule Dictionary" of fuzzy logic atoms.

2. Intermediate ANFIS Construction via Group Rule Interpolation

This stage is the core innovation of the research. Using Locally Linear Embedding (LLE), a manifold learning algorithm, the system identifies and interpolates a group of relevant rules from the dictionary to cover the target domain. The underlying assumption is that if the conditions in the target domain can be represented as a linear combination of conditions in the source domains, the corresponding results should follow a similar linear pattern.

Illustration of the group rule interpolation process using antecedent and consequent rule dictionaries.

3. Fine-Tuning

The interpolated rules form an "intermediate" ANFIS. While already functional, this model is further refined using the limited target data available. This process employs standard backpropagation and least-squares estimation to optimize the network parameters, ensuring the model is tailored to the specific nuances of the sparse target dataset.

Experimental Results

The effectiveness of the proposed method was validated across synthetic functions, 12 real-world regression datasets from the KEEL repository (e.g., Diabetes, AutoMPG, Concrete strength), and image super-resolution tasks.

  • Error Reduction: On real-world benchmarks, the proposed method achieved an average Root-Mean-Squared Error (RMSE) of 3.189, a significant improvement over the 6.060 RMSE produced by a standard ANFIS trained only on the sparse data.
  • Stability: The interpolated models demonstrated lower standard deviation in performance, indicating higher reliability and less sensitivity to the specific distribution of the sparse training points.
  • Visual Fidelity: In image super-resolution, the method effectively reduced artifacts and "bad edges," outperforming standard TSK-type models in reconstructing high-resolution details from limited information.

Conclusion

By transitioning from single-rule interpolation to group-based system construction, this research provides a robust solution for building fuzzy models in data-poor environments. The application of manifold learning to fuzzy rule sets allows for a sophisticated mapping of logic across domains. Future work will explore more advanced automated clustering techniques and the potential for iterative interpolation to further enhance model complexity and accuracy.

Summary updated

Abstract

A major assumption for constructing an effective adaptive-network-based fuzzy inference system (ANFIS) is that sufficient training data are available. However, in many real-world applications, this assumption may not hold, thereby requiring alternative approaches. In light of this observation, this article focuses on automated construction of ANFISs in an effort to enhance the potential of the Takagi-Sugeno fuzzy regression models for situations where only limited training data are available. In particular, the proposed approach works by interpolating a group of fuzzy rules in a certain given domain with the assistance of existing ANFISs in its neighboring domains. The construction process involves a number of computational mechanisms, including a rule dictionary which is created by extracting the rules from the existing ANFISs; a group of rules which are interpolated by exploiting the local linear embedding algorithm to build an intermediate ANFIS; and a fine-tuning method which refines the resulting intermediate ANFIS. The experimental evaluation on both synthetic and real-world datasets is reported, demonstrating that when facing the data shortage situations, the proposed approach helps significantly improve the performance of the original ANFIS modeling mechanism.
BibTeX
@article{yang2019anfis,
  author = {Jing Yang and Changjing Shang and Ying Li and Fangyi Li and Qiang Shen},
  title = {{ANFIS Construction With Sparse Data via Group Rule Interpolation}},
  journal = {IEEE Transactions on Cybernetics},
  year = {2019},
  doi = {10.1109/tcyb.2019.2952267},
}
Figure