A Learning-Based Adjustment Model with Genetic Algorithm of Function Point Estimation

Abstract

Software effort estimation is a crucial activity in the field of software engineering. Vast related research has been conducted and Function Point Analysis (FPA) is widely used among various estimation methods and models. Traditional methods derived from function points are good sizing but do not reflect everchanging software technology and development process. This paper aims to propose a learning-based adjustment model for FPA by applying Genetic Algorithm (GA) and analogy-based estimation. Analogy-based method is used to select Key Influence Factors (KIFs) of software effort and related adjustment coefficients in a multiplicative equation, and the optimized process is driven by GA. A case study is presented to train the process with a suitable model by using ISBSG dataset, and the trained model is applied to estimate new projects collected by the same organization and a real completed project. The evaluation result shows that applying a suitable multiplicative model to adjust original function points is a feasible approach to improve accuracy of software effort estimation. It also demonstrates that the proposed model is comparable with other effort estimation methods, including IFPUG method and analogy-based estimation.