Please use this identifier to cite or link to this item:
Title: Evaluation of Feature Selection Techniques for Software Maintenance Prediction
Authors: Nanda, Sheena
Supervisor: Saxena, Sharad
Bala, Anju
Keywords: Feature Selection;Machine Learning;Software Selection
Issue Date: 11-Aug-2017
Publisher: Thapar University
Abstract: Software quality is the ease with which a process or software fulfills customer’s expectation. One of the essential steps in measuring the quality of the software is software maintenance. It has been found that software maintenance accounts for 60-70% of the total cost thus, it is very important to prepare an accurate software maintenance plan during the software development. This helps to analyze the cost and risk associated with the software well in advance so that an optimized resource planning can be done. The development of a software is associated with high variance on techniques and goals; therefore, it is hard to measure software’s quality. Hence, object-oriented metrics have turned into an essential part of software development process for quantitative measurement. However, not all the metrics affects the quality in same way. Feature selection methods decreases computation time by better understanding the data using machine learning. Hence, the objective of this dissertation has been to construct improved prediction model by pre-processing the data using feature selection techniques. Feature selection algorithms decrease the input variables by selecting the most relevant ones and removing the redundant variables. In this dissertation, `CHANGE' is used as the measure of maintainability. Two open source softwares: `Apache Log4j' and `Drumkit' have been considered for this research work and input variables have been extracted using CKJM and LocMetric tools. Further, JarComp tool has been used to calculate CHANGE per class among two versions of the open source software systems. Thereafter, most relevant features were extracted using seven different feature selection techniques and further, their effectiveness was measured using nine machine learning algorithms for software maintenance prediction. Subsequently, different combinations of feature selection techniques and prediction models have been analyzed to identify the best combination. This dissertation will enable software developers to predict the change requirements in advance using software metrics that are relevant and affect the efficiency of the software which further helps to improve the quality of software thus, encouraging good coding and designing techniques
Description: Master of Engineering -CSE
Appears in Collections:Masters Theses@CSED

Files in This Item:
File Description SizeFormat 
4628.pdf4.13 MBAdobe PDFThumbnail

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.