Attribute reduction in incomplete information system based on rough set theory using fuzzy imperialist competitive algorithm

چکیده مقاله

The paper "Attribute Reduction in Incomplete Information System based on Rough Set Theory Using Fuzzy Imperialist Competitive Algorithm," published in the Journal of Information Technology Management in 2017, addresses the challenge of feature selection in datasets with missing values, known as incomplete information systems. Authored by Mohammad Ghanei Ostad, Hosein Khosravi Mahmoee, and Majid Abdolrazzagh Nezhad, the study acknowledges that classical rough set theory, while powerful for knowledge discovery, is not directly applicable to incomplete data. The researchers propose a novel hybrid approach that integrates fuzzy logic with the imperialist competitive algorithm (ICA) to perform attribute reduction without imputing or altering the missing values, treating the incomplete system as a complete one for analytical purposes.

The core innovation lies in the enhancement of the ICA through fuzzy logic to dynamically control its key parameters, such as the revolution rate, assimilation coefficient, and imperialist power. This fuzzy-ICA adapts more intelligently to the search landscape, leading to improved convergence and solution quality compared to the classic ICA. The method employs rough set theory to evaluate the dependency and significance of attribute subsets, aiming to minimize the number of features while preserving informational integrity. The fuzzy rules were designed based on preliminary tests, allowing the algorithm to adjust its behavior in response to the state of the search process, thereby reducing the risk of local optima and enhancing exploration.

Experiments were conducted on five real-world UCI datasets with missing values, including Lung Cancer, Dermatology, Wisconsin Breast Cancer, Hepatitis, and Flag. The proposed fuzzy-ICA was compared against the classic ICA and a genetic algorithm. Results demonstrated that the fuzzy-ICA consistently achieved better or equal reducts (minimal feature subsets) in most cases, and for several datasets, it even outperformed the best-known solutions reported in prior literature. The fuzzy version showed superior convergence behavior, lower variance in outcomes, and a more stable performance across multiple runs, indicating its robustness and efficiency in handling incomplete data.

The key achievement of this research is the successful integration of fuzzy logic with a metaheuristic algorithm to tackle a complex, discrete optimization problem in data preprocessing. The fuzzy-ICA not only improves solution quality but also offers a dynamic, self-tuning mechanism that reduces the need for manual parameter adjustment. This approach provides a practical and effective tool for feature selection in real-world scenarios where data incompleteness is common, thereby aiding in more efficient knowledge discovery and pattern recognition. The study also sets a new benchmark for attribute reduction on the tested datasets, showcasing the potential of hybrid intelligent systems in data mining applications.

In conclusion, the paper presents a significant step forward in handling incomplete information systems by combining rough set theory with an adaptive fuzzy-based imperialist competitive algorithm. The positive outcomes highlight the value of incorporating soft computing techniques to enhance metaheuristic search processes. Future work may explore extending this framework to larger and more complex datasets, or integrating other intelligent mechanisms to further improve scalability and accuracy in attribute reduction tasks.