Fuzzy Analogy Based Effort Estimation: An Empirical Comparative Study

Abstract

Software Development Effort Estimation (SDEE) plays a primary role in software project management. Among several techniques suggested for estimating software development effort, analogy-based software effort estimation approaches stand out as promising techniques.In this paper, the performance of Fuzzy Analogy is compared with that of six other SDEE techniques (Linear Regression, Support Vector Regression, Multi-Layer Perceptron, M5P and Classical Analogy). The evaluation of the SDEE techniques was performed over seven datasets with two evaluation techniques (All-in and Jackknife). The first step of the evaluation aimed to ensure that the SDEE techniques outperformed random guessing by using the Standardized Accuracy (SA). Then, we used a set of reliable performance measures (Pred(0.25), MAE, MBRE, MIBRE and LSD) and Borda count to rank them and identify which techniques are the most accurate.The results suggest that when using All-in evaluation, Fuzzy Analogy statistically outperformed the other SDEE techniques regardless of the dataset used. However, when using Jackknife evaluation, the results obtained depended on the dataset and the SDEE technique used. The results suggest that Fuzzy Analogy is a promising technique for software development effort estimation.