SOFTWARE COST ESTIMATION MODEL BASED ON NEURAL NETWORKS

Software engineering is providing estimation models in order to evaluate the costs for important phases of software development such as requirements analyses, high and low level design, development, testing, deployment and maintenance. The paper proposes a software cost estimation model based on neural networks as an alternative for traditional cost estimation models. The paper argues that there are at least two considerable advantages for the neural network model: neural networks can learn from previous experience and secondly, neural networks can discover a complete set of relations between dependent variables (software cost, software size, required effort, etc.) and independent variables (cost drivers). Experiments and simulations are made in order to prove the theoretical claims.