Distributed fuzzy rule miner (DFRM)
Samane Sharif, Mohammad Reza Akbarzadeh-T · 2013
Nowadays scalability and capability of parallel execution are the most important characteristics for data mining algorithms due to the growing size of data sets. In this paper, a new distributed framework called DFRM is proposed to extract fuzzy rules from numerical data using a multi-agent approach. These extracted rules can be used for classification and decision making tasks. Scalability, self-organization and uncertainty handling are important characteristics of the proposed system. Scalability and self-organization are provided by autonomous agents in the learning process. Interaction among agents can lead to a more compact fuzzy rule base for decision making. Moreover the training samples are split equally between the agents randomly. Therefore each agent has a partial view of data set. Four UCI data sets are used to evaluate the proposed framework based on accuracy and rule base size. Experimental results show that the resulting distributed classification model maintains acceptable accuracy with fewer rules In addition, this model is robust against non-availability of training data.