Measuring object-oriented software maintainbility in design phase using structural complexity and aesthetic metrics

Abstract

Qualities of software design heavily affect on qualities of software ultimately developed. One of the claimed advantages of object-oriented paradigm is the ease of maintenance. Assessing maintainability in design phase will help software designers to decide that the design of software should be altered or not. This will help reducing cost of software maintenance in later phases. This research presents a controlled experiment carried out to investigate whether structural complexity and aesthetic metrics can be indicators of class and sequence diagrams maintainability and to establish maintainability models from both metric sets. Maintainability models are constructed by using three techniques: Discriminant analysis, Decision tree and MLP neural network in order to find the best one among these models. The preliminary result shows that the maintainability models constructed by applying MLP neural network are the best models in terms of the model accuracy. Moreover, this research constructs an automated tool for predicting maintainability using the maintainability models and also proposes a new set of structural complexity metrics for measuring maintainability from class diagrams.