Quality Model for Analyzability Evaluation Of Java Class

Abstract

We propose a quality model for evaluating the analyzability of java class. Our model determines how difficult to understand the code in order to identify cause and error locations. To create this prediction model, we survey software metrics likely to have an impact on class’s analyzability. We applied the ordinal logistic regression method to identify a set of metrics correlated with the analyzability level divided into 3 levels: poor, fair and excellent. To validate our model, we tested it at 95% confidence level using a set of 37 java classes from jEdit open source program. We found that only Coupling between Object (CBO) and Lack of Cohesion in Methods (LCOM) metrics influence class analyzability.