Aspect mining using Self-Organizing Maps with method level dynamic software metrics as input vectors

Abstract

A major impediment to program comprehension, maintenance and evolvability is the existence of crosscutting concerns scattered across different modules tangled with implementations of other concerns. Presence of crosscutting concerns in software systems can lead to bloated and inefficient software systems that are difficult to evolve, hard to analyze, difficult to reuse and costly to maintain. This paper shows that clustering based on easily extractable software features derived through method calls, parameter sharing and method interactions represented as dynamic metrics can be used to determine code scattering and or tangling, thereby providing a steppingstone towards identifying crosscutting concerns leading to mining of possible aspect candidates. A three-step approach is used in the Aspect Mining methodology introduced in this paper. In the first step, two legacy programs were dynamically traced, and data representing interaction between code fragments were collected. In the second step, metrics were formulated from the collected data and submitted as input to Self Organizing Maps for clustering. In the third step, the obtained clusters were mapped against the test programs in order to identify code scattering and tangling symptoms, leading to identification of aspect candidates. Finally viable validation methodologies were applied to assess performance, and establish the validity of the methodologies used. Results obtained in this paper are found to have matched or exceeded results obtained in other existing Aspect Mining methods

On Automatic Identification of Monitoring Concerns Implementation

“…), and are called automated aspect mining techniques. Different approaches are used: clustering [9][10][11], clone detection techniques [12][13][14], metrics [15], association rules [16], formal concept analysis [17,18], execution relations [19,20], self organizing maps [21], and link analysis [22].…”

Feature selection for clustering based aspect mining

“…The feature models are Method spread (MSP) [3], Fan-in value (FIV) [1-3, 6, 15], Fan-out value (FOV) [3], method internal coupling (MIC), method external coupling (MEC) [3], affected classes (AC) [3,15], Henry and Kafuras structure complexity (C p ) [18], return value (RV) [7], the number of parameters (NPM) [20], method signature (MSig) [3] and cohesion on method (COM) [20].…”

“…Applying the clustering technique, the crosscutting concerns would be classified into the similar groups and other non-crosscutting concerns are divided into other groups. A lot of prior research [1][2][3] focus on using clustering techniques to achieve aspect mining.…”

“…Several studies [1][2][3] propose to use clustering techniques for mining aspects. However, how to get the optimal set of metrics [1,3,4], remains open, which affects the performance of clustering. The metrics are called feature models in this paper and the set of values of metrics is features.…”

Using Software Metrics for Automatic Software Design Improvement

“…In [18], the fan-in and the fan-out of a method is presented. The definition says that fan-in is the number of distinct method bodies that invoke a method and the fan-out is the number of times a method invokes another methods.…”