Dynamic Density-Based Fuzzy Rule Interpolation with Application to Mammography Abnormality Detection

Fuzzy rule interpolation (FRI) offers a powerful solution for deducing conclusions when observations fail to match existing rules in a sparse fuzzy rule base. Traditional fuzzy rule-based systems often face challenges from insufficient data or a lack of comprehensive human expertise to define rules covering the full problem domain. Starting from a sparse rule base, FRI generates interpolated rules to address unmatched observations. However, the interpolated rules, which may contain valuable insights about the problem space, are typically discarded after use, limiting the system's potential to improve efficiency and robustness. This calls for dynamic fuzzy rule interpolation, where interpolated rules are evaluated and promoted to the rule base for future use, thus expanding domain coverage and enhancing inference efficacy. This paper introduces a dynamic density-based fuzzy rule interpolation iteration framework. It combines the popular transformation-based fuzzy rule interpolation with the effective method of ordering points to identify the clustering structure hidden in the interpolated outcomes. By dynamically enriching the rule base, this approach enables fuzzy rule-based systems to acquire self-learning capabilities. Experimental results on real world dataset demonstrate the effectiveness of the proposed method in enhancing system performance and robustness, in addressing the real-world problem of mammography abnormality detection.