Software Quality Prediction Using Data Mining Techniques

The development of software systems is hard process. The software engineers doing their best during the software construction levels to produce high quality systems. Quality measured from multi perspectives. External and internal perspectives determined by McCall’s quality model. In the scope of our work, we concerned with the internal quality. The code level represents the internal quality of the software. By using object-oriented metrics, we can measure code quality. These metrics reflects a clear idea about the code under investigation. Therefore, we proposed to use them during the code test level of the project. If there are any outliers in the values of the selected metrics, we can revise them by applying the appropriate corrections at the code. The code has variant design problems called design smells or code smells. These problems affects directly on the future maintenance of the project. One of the widely spread problem is the Feature Envy method. This type of code design problems causes complicated maintenance problems. In order to make the maintenance process completed in optimum time we have to get rid of this code problem. Therefore, earlier detection is the right way to minimize the time needed for expected maintenance and development. In this paper, we combined the OO metrics with some data mining algorithms to classify the code’s methods into two types as Feature Envy or not. The paper applied the technique on an open source java project. In addition, evaluate the results with other selected detection tools with different techniques. The results registered a high accuracy in the case of using cross validation against using training set option in K-Nearest neighbor algorithm.