Ripple Effect Identification in Software Applications

Introduction

Software systems undergo numerous changes throughout their life span. Rectification and enhancements are two main factors responsible for these changes made to software applications. Software maintenance is a very important phase in the development life cycle and it has been observed that it employs the lion’s share of the total allocated expense. Researchers have found that ripple effect is capable of picturing the extent of change effect on the rest of the software application (Arvanitou, Ampatzoglou, Chatzigeorgiou, & Avgeriou, 2015). Black (2006) have studied the role of ripple effect in software evolution and they found it very beneficial in the software maintenance phase. “It can highlight modules with high ripple effect as possible problem modules which may be especially useful in preventive maintenance. It can show the impact in terms of increased ripple effect during perfective and adaptive maintenance where the functionality of a program is being modified or its environment has changed,” according to Black (2006).

Ripple effect identification aims to classify those classes that are modified as aftereffects of changes made to a software application. These change prone classes demand allocation of additional resources like labor, budget and time for their thorough testing and meticulous tracking. Good planning and allocation help in minimizing errors leads to good quality software. Thus, it helps in better optimization and proficient exploitation of prevailing resources.

In this paper, we propose development of quality models by establishing the correlation amongst object-oriented (OO) metrics and ripple effect of pair of classes. The proposed model is developed using learning techniques. Different learning techniques (statistical, machine learning [ML] or Search Based Techniques [SBT]) are available to our disposal. The selection of effective learning technique is a major challenge. Agrawal and Singh (n.d.) studied the correlation amongst OO software metrics and ripple effect recently. Prediction model was derived using statistical technique in this work. Performing a thorough search of relevant literature, we found that the potential of other learning techniques such as ML and SBT has not been explored yet for ripple effect identification studies.

ML applies instructed procedures that accept and examine entered data to forecast output values within limits. When new data is provided as input to these procedures, they learn and adjust their functions for improved results and thus become intelligent with the passage of time. SBT are one of the most emerging practices in the recent years. They perform iterative search across the complete solution universe and identify the optimum or near optimum solution. It starts with set of candidate solutions and calculate the fitness for each candidate. Mutation, crossover, reproduction or other selection techniques are then applied to select the best candidate for future generations (Eiben & Smith, 2003). Until the desired quality of results is obtained, the algorithm goes on generating future generations. Several studies in literature support the usefulness of SBT to solve software engineering problems that have certain inconsistencies, no perfect answer or no pinpoint rules for effective solutions (Harman & Jones, 2001). The ripple effect identification for change prone classes reveals similar behavior. There are no precise rules and at times insufficient data when identifying ripple effect of changed classes in software applications.

We study the efficiency of the statistical technique (Logistic regression), five ML and seven Search Based Techniques for ripple effect identification and compare their results using non-parametric test. We evaluate the ten-fold validation results for validating the proposed prediction models developed with different learning techniques. The following research questions are explored in this paper:

• RQ1: Are the OO metrics of classes associated with the ripple effect?

• RQ2: Can we use SBT and ML techniques for ripple effect identification problems?

• RQ3: How is the relative performance of different techniques for ripple effect identification of changed classes?