Mining classification rules from fuzzy min-max neural network

The basic fuzzy min-max neural network (FMMN) is capable to perform the supervised classification of data. As like other artificial neural networks, FMMN is also like a black box and expressed in terms of min-max values and associated class label. So the justification of classification results given by FMMN is required to be obtained to make it more adaptive to the real world applications. This paper proposes the model to extract classification rules from trained FMMN. These rules justify the classification decision given by FMMN. For this FMMN is trained for the input data and resulting min-max values in the quantized form and class labels of created hyperboxes are passed as a input to the partial rule extraction (PART) algorithm of Waikato Environment for Knowledge Analysis (WEKA). As a result, more comprehensible rules expressed in terms human readable linguistic terms are extracted. The proposed model is applied to iris and wine dataset taken from the UCI machine learning repository. Thus the extracted rules represent the trained FMMN in the more understandable form and can easily be applied to real world applications.