Software Measurements from Machine Learning to Deep Learning

Abstract

Software measurement (SM) is an umbrella activity during the entire software development cycle. Measurements and metrics of the attributes are indispensable for successful completion of project and effective delivery of software product. This chapter discusses SMs using deep learning (DL) techniques from the perspective of an empirical study. It is evident that an inaccurate prediction or estimation during the software development processes leads to loss of money and loss of projects. Since the beginning of software engineering, a wide range of methods are being deployed for measuring the software attributes. At present, the conventional techniques are not so apt for SMs due to excessive complex attributes of very large software. Machine learning (ML) has been the answer to all market needs in the past 30 years. It is noticed that ML is quite good to perform measurements in software engineering processes, but it is not the best method and needs enhancements. DL is the extension to ML, which is now being extensively used for SMs. The chapter begins with an introduction to ML and DL techniques and their applications in SMs empirically. Then, it highlights the literature work carried out in the field of empirical SMs using DL techniques. One of the most important DL techniques is convolutional neural networks which is discussed as a case study. This study provides a practical orientation to the readers about the implementation of DL technique to SMs.