Abstract-Software cost estimation is an important phase in software develop ment. It predicts the amount of effort and develop ment time required to build a software system. It is one of the most critical tasks and an accurate estimate provides a strong base to the development procedure. In this paper, the most widely used software cost estimat ion model, the Constructive Cost Model (COCOM O) is discussed. The model is implemented with the help of artificial neural networks and trained using the perceptron learning algorithm. The COCOMO dataset is used to train and to test the network. The test results fro m the trained neural network are co mpared with that of the COCOMO model. The aim o f our research is to enhance the estimation accuracy of the COCOMO model by introducing the artificial neural networks to it.
Human effort is one of the main resources of software cost estimation. A successful software project development primarily relies on accurate effort prediction at an early stage of development. There are many effort prediction models in the literature. Deciding which model to choose is a challenge for the project managers. This paper investigates whether it is possible to improve the accuracy of software cost estimations by coupling firefly algorithm with the existing artificial neural network (ANN) models used in software cost predictions. The firefly algorithm is one of the recent evolutionary computing models inspired by the behaviour of fireflies in nature. This is compared with particle swarm optimization used already in literature for software cost estimations. The ANN models examined in this work include radial basis function network and functional link artificial neural networks models. The experimental results show that ANN models perform extremely well by incorporating firefly algorithm and intuitionistic fuzzy C-means for data preprocessing. The proposed approach is empirically validated through a statistical framework.All the software effort estimation studies are divided into two general classes, sparse data and many data methods. The sparse data methods require few or no historical data such as expert judgement techniques, where the estimate is produced based on judgmental processes of a software expert. The many data methods are based on available datasets and may be subdivided into parametric, nonparametric and semiparametric [4,5].Parametric models clearly describe the underlying relationship among parameters through an equation. Here, the predictors are linearly associated with the response variables. These models are applied initially for software cost estimation [6][7][8][9][10][11][12][13]. The non-parametric models are more robust. They do not depend on any specific data distribution and nor assume any fixed structure for a model. Thus, these models have many advantages then its parametric counterpart. The semiparametric models contain the merits of both parametric as well as non-parametric models. They are used especially in cases where there is less information on the relationships between the dependent and the independent variables. They can identify complex relations between variables for which there is no predetermined model structure or in cases where non-parametric model fails [5].This work focuses on non-parametric studies especially artificial neural network (ANN) models for software cost estimation. The ANN models used in our approach are functional link artificial neural networks (FLANN) and radial basis function network (RBFN). The reasons for choosing them are RBFN has the advantages of easy design, good generalization, strong tolerance to input noise and online learning ability [14], whereas FLANN also can beautifully handle the data, which are highly nonlinear in nature. It has faster convergence and has less computational complexity as is associated with multilayer neural net...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.