Artificial Bee Colony-Trained Functional Link Artificial Neural Network Model for Software Cost Estimation

Abstract

Software cost estimation is forecasting the amount of developmental effort and time needed, while developing any software system. A good volume of software cost prediction models ranging from the very old algorithmic models to expert judgement to non-algorithmic models have been proposed so far. Each of these models has their advantage and disadvantage in estimating the development effort. Recently, the usage of meta-heuristic techniques for software cost estimations is increasingly growing. So in this paper, we are proposing an approach, which consists of functional link ANN and artificial bee colony algorithm as its training algorithm for delivering most accurate software cost estimation. FLANN reduces the computational complexity in multilayer neural network, and does not has any hidden layer, and thus has got fast learning ability. In our model, we are using MRE, MMRE and MdMRE as a measure of performance index to simply weigh the obtained quality of estimation. After an extensive evaluation of results, it showed that training a FLANN with ABC for the problem of software cost prediction yields a highly improved set of results. Besides this, the proposed model involves less computation during its training because of zero hidden layers and thus is structurally simple.