Software Package Testability Prediction Using Object-Oriented Cohesion Metrics

Software development costs can be greatly reduced by creating well-designed software. Object-oriented software metrics can assist in designing high-quality systems. Package cohesion metric is found to be highly related to software testability. We have developed a package cohesion metric that has proved its ability in software maintainability and testability predictions as early as the design phase of software development. In this paper, we expand our previous works by proposing a logistic regression model that can predict software testability regarding its costs and efforts. The proposed model can predict the package testability during the design phase, which can save many resources of software development life. This early prediction helps developers to discover the poorly designed packages and take decisions to reduce any further costs and efforts that might happen in the future. The experiments showed that the proposed package cohesion metric could assess the design cohesiveness of software packages.

I. Introduction

Software testing is one of the software development process activities. It is a fundamental activity where faults and errors are detected to ensure the quality of the software system. Testability refers to the ease of discovering these errors and faults. Not all software systems have the same level of testability [1]. As known, software testing is one of the software maintainability characteristics and both are highly related. The high maintainable the software is, the high it will be testable. Software testing aims at delivering high-quality software systems, and software testability can facilitate this process [2]. Software testability is defined as "attributes of software that bear on the effort needed to validate the software product," as defined by ISO [3]. Another definition by ISO is the degree of efficacy and efficiency with which test criteria for a system can be defined and tests to assess whether those criteria have been satisfied [4]. IEEE defines software testability as the degree to which software supports the creation of test criteria and the execution of tests to determine if those requirements have been met [3]. All these definitions share the same goal of software testability that links it to reduced testing efforts and higher software quality. Whereas the late diagnosis of a lack of testability can be difficult and costly to repair, it can have a negative impact on the testing and maintenance effort [5].