Abstract
One of the challenges in the analogy-based approach is the identification of relevant project features to be used in its application [2]. Related literature reveals that a number of search algorithms are currently in use for selecting useful project features, but none of these algorithms provide useful statistics to determine the quality and the relevance of the features selected by using one of the search techniques. In most cases, features are selected using brute-force and other forms of search heuristics, which requires a large amount of computing power and time to complete its entire search function, yet provides no evidence to support its identified features for estimation purposes. This is especially problematic when dealing with large datasets. Furthermore, software development duration estimation is something of an afterthought. For example, in COCOMO duration is estimated using a model that uses effort as an input. In practice, the model would be driven by the effort estimate obtained from the effort estimation model [3]. Using regression analysis, researchers have proposed estimating duration using either standard product factors or effort estimates [4]. Putnam's SLIM model is a notable exception, since it is based on a model that relates size to effort and duration [5]. We have developed a solution called Analogy-X in [6, 7] to overcome many of these issues when selecting relevant project features based on Mantel's correlation statistics and randomization tests. In this paper, we apply Analogy-X on a much larger publicly available dataset to evaluate its performance and its capability in dealing with outlying data points. Also we argue that it is possible to determine a set of project features that are appropriate for joint effort and duration prediction. We demonstrate this on a company dataset where duration and effort are dependent variables. Our method can also identify whether effort and duration should be estimated using different feature sets. However, if this is the case, we must caution estimators (or practitioners) to be aware that We developed a novel method called Analogy-X to provide statistical inference procedures for analogybased sof