An activeness metric for the quality of design of a Multilayer Perceptron Model

Abstract

The objective of this research work is to verify the capability of a metric to measure the activeness of a Multilayer Perceptron (MLP) Model in order to determine the quality of design of the MLP. The analysis of this metric will indicate the necessary modifications which need to be made to an MLP to improve its performance. The activeness of a neural network can therefore be defined as the degree of readiness of a neural network to respond to inputs by giving outputs which are as accurate as possible. This definition is similar to those of other activeness metrics as defined in [3–6]. An existing software metric, that is, the Root Mean Square Error (RMSE), is commonly used to measure the quality of performance of an MLP. However, this metric gives very limited information for measuring the quality of design of an MLP. The research work involves the development of a new software metric called Neural Network Quality of Design Activeness Metric (NNQDAM) which will be able to gauge the quality of design of an MLP. The metric will take into account several parameters, including the number of neurons in each of input, hidden and output layers and a summarized value of weights of connections between neurons, which will help to identify the quality of design so as to improve the performance of an MLP.