Multi-Agent System Based Data Mining Technique for Supplier Selection

Different target knowledge representations, which correspond to different hypothesis spaces, are suitable for learning various types of target functions. A distinct learning algorithm, capitalizing on an underlying structure to guide the exploration of the hypothesis space, leverages each of these hypothesis representations. Consequently, making decisions on these matters entails navigating an expansive landscape of alternative approaches to determine the one that best aligns with the defined learning problem. To select an appropriate learning algorithm that performs optimally for the given problem and target function, it is essential to analyze the interplay between the size and completeness of the hypothesis space, the availability of training examples, the learner's prior knowledge, and the confidence in a hypothesis's ability to generalize correctly to unseen data points. This challenge often arises in the context of poorly integrated systems. The objective of this paper is to introduce a specific approach to addressing these challenges, involving the use of enhanced elicitation techniques and tools, improved development paradigms, knowledge modeling languages, ontologies, and advanced methods for system maintenance. However, in the past decade, an alternative reasoning paradigm and computational problem-solving approach have garnered significant attention, involving the use of Multi-Agent System (MAS) based Data Mining (DM) techniques. This paper proposes the integration of MAS and DM for supplier selection, with a focus on partitioning-based algorithms such as K-means, Farthest First, Expectation Maximization, and Filtered Clusterer. These algorithms are compared based on factors such as dataset size, the number of clusters, and the time required for cluster formation. Finally, within this integrated model, MAS-DM is exemplified through an illustrative result-matching process.