Neural Computing and Applications Software
Optimizing colonial competitive rule-based classifier using a hybrid approach
Abstract
Classification algorithms are widely applied across various domains, and they are commonly categorized into types such as instance-based, neural network-based, and rule-based classifiers. Colonial Competitive Rule-based classifier is a rule-based classifier that leverages the Imperialist Competitive Algorithm as its core engine. It extracts rules by evaluating all features of the dataset simultaneously. While Colonial Competitive Rule-based classifier has shown promising results compared to several standard classifiers in previous studies, this paper introduces a novel hybrid method to further enhance its classification accuracy. Our proposed approach incorporates two key contributions. First, instead of using randomly initialized rules, we initialize Colonial Competitive Rule-based classifier with rules generated by the Repeated Incremental Pruning to Produce Error Reduction algorithm, which improves the initial rule quality. Second, we employ a fuzzy Mamdani inference system to dynamically adjust one of Colonial Competitive Rule-based classifier’s crucial hyperparameters over time, thereby improving adaptability. To assess the effectiveness of our proposed method, we performed experiments on nine real-world datasets and analyzed the results against those of existing techniques. The experimental findings indicate that the enhanced Colonial Competitive Rule-based classifier achieves consistently better accuracy and outperforms both the original Colonial Competitive Rule-based classifier and several other classification models.