Changeability prediction model for java class based on multiple layer perceptron neural network
Songsakdi Rongviriyapanish, Thanapol Wisuttikul, Boonchai Charoendouysil, Pattarin Pitakket, Pattanan Anancharoenpakorn, Panita Meananeatra · 2016
A quality model for assessing the changeability level of java code is important for software development. It permits developer to know which classes to be improved for having a better software maintainability. Moreover, a good quality model must be created based on a set of well-selected attributes and metrics. Currently, no research work proposes a changeability assessment model that takes into consideration the metrics covering ten relevant object-oriented attributes. We propose a class changeability prediction model developed by using the multilayer perceptron (MLP) as a classifier method and a training data set of 137 java classes from jEdit open source project for training the model. Model accuracy attains 89.81% and the model can perfectly separate java classes with good changeability level from those with poor or fair changeability levels.