Rule Extraction from a Neural Network with Segregated Numeric and Categorical Input

R.K. Brouwer (Canada)


Soft computing, categorical variable regression, feed forward neural networks, indicator variables


The data on which a MLP (multi-layer perceptron) is to be trained to approximate a continuous function may include inputs that are categorical rather than only numeric or quantitative. However an MLP with connection matrices that multiply input values and sigmoid functions that further transform values represents a continuous mapping in all input variables. A MLP therefore requires that all inputs correspond to numeric, continuously valued variables and represents a continuous function in all input variables. On the other hand a categorical variable produces a discontinuous relationship between an input variable and the output. The way that this problem is often dealt with is to replace the categorical values by numeric ones and treat them as if they were continuously valued. However there is no meaningful correspondence between the continuous quantities generated this way and the original categorical values. An approach examined in this paper is to train a standard feedforward network with the numeric portion of the input only and train another network with the categorical portion of the input to produce a selector vector. The selector vector produced by the second network selects output from the first network.

Important Links:

Go Back