Paper Title
Improving The Performance of Neural Networks for Software Complexity Measurement using Steepness Coefficient

Abstract
Software complexity measurement is an essential task for development and management of software projects. In the literature, there are some software metrics to measure the complexity of software. However, instead of using conventional methods for obtaining software metrics, introducing novel estimation models for that purpose is needed . This paper presents results for estimation of software complexity metricsby neural networks in which the steepness coefficient (s) in the activation function is applied. We use a neural network of three layers with a single hidden layer and train this network by using Scaled Conjugate Gradient(SCG) Algorithm. We apply sigmoid activation functionusing various steepness coefficients (s) to determine the effects of this parameter on the performance of neural network model. Experiments performed on a widelyknown software metrics, Halstead, to indicate that neural networks approach is feasible. We compare our results of software complexity obtained by using neural networks with those calculated by Halstead model. This comparison shows that, for software complexity measurement, an optimized value of the steepness coefficient improves the performance and the effectiveness of our proposed method of neural networks. Keywords- Software Complexity, Neural Networks, Steepness Coefficient