Comparison and assessment of improved grey relation analysis for software development effort estimation

The goal of software project planning is to provide a framework that allows project manager to make reasonable estimates of the resources. In fact, software development is highly unpredictable - only 10% of projects on time and budget. Thus, it is very important for software project managers to accurately and precisely estimate software development effort since the resources are limited. One of the most widely used approaches of software effort estimation is the Analogy method. Since the method of Analogy is constructed on the foundation of distance-based similarity, there are still some drawbacks and restrictions for application. For example, the anomalistic and outlying values will influence the function to determine similarity. Contrarily, Grey Relational Analysis (GRA) is a distinct measurement from the traditional distance scale and can dig, out the realistic law from small-sample data. In this paper, we will show how to apply GRA to evaluate the effort estimation results for different data sequences and to compare its accuracy with that of Analogy method. Experimental result shows that the GRA provides a better predictive performance than other methods. We can see that the GRA is more suitable for predicting software development effort with unbalanced dataset.