Contents I. Introduction
A Software Product Line (SPL) is basically a collection (family) of conceptually alike software systems. These conceptually similar products are also called product variants [1]. They share some common features (commonality) with other variants of the family and also have unique features (variability) of their own. This commonality and variability is often described with the help of features. Features are abstract entities which are later mapped and implemented through artifacts like files, codes, components etc. The basic purpose of features is to depict functionality of an SPL. They help users in understanding and selecting variants, and also assist developers during engineering and evolving an SPL. Unfortunately, quality assurance (QA) of features in later stages of SPL development is challenging as well as costly. At the time of coding, testing features in isolation is difficult, because they get scattered across the entire codebase and the only way to test them is during integration time [2] [3] [4] [5] [6]. Whereas early assessment of feature models (during their design time) will ensure high level quality in the preliminary phase. Prediction models can also reduce the quality assessment costs during the later stages. A number of studies have investigated feature models during the implementation phase of feature models. These studies have achieved satisfactory results as well. But these models cannot be used for quality assessment at the feature level. Early assessment of feature model usability will (i) help developers in selecting (or prioritizing) features in such a way that maximum usability can be attained for the product line, (ii) ensuring future usability as well as possible reusability and utility of the product line. (iii) Improvise the actual quality assessment at the feature model level rather than assessing quality at later stages. As usability is an external attribute and can be assessed only during the later stages, we have identified three sub characteristics viz. learning ability, understandability and communicativeness which can be assessed in the early phase of designing. In the current research, we propose to develop prediction models for each of these usability sub characteristics with the help of a subset of metrics. Based on literature survey, 5 machine learning algorithms are selected for developing the prediction model. The best performing prediction model will be further trained over a set of feature models. The prediction will be cross verified by obtaining user’s view over usability on the same set used for testing. Post the training it will be tested and prediction will be done on a new set of feature models. Both the values obtained will be compared. Section II discuses the design of the study, and section III contains the conclusion.
