Assessing the Effectiveness of Defect Prediction-based Test Suites at Localizing Faults

Abstract

Debugging a software program constitutes a significant and laborious task for programmers, often consuming a substantial amount of time. The need to identify faulty lines of code further compounds this challenge, leading to decreased overall productivity. Consequently, the development of automated tools for fault detection becomes imperative to streamline the debugging process and enhance programmer productivity. In recent years, the field of automatic test generation has witnessed remarkable advancements, significantly improving the efficacy of automatic tests in detecting faults. The localization of faults can be further optimized through the utilization of such sophisticated tools. This dissertation aims to conduct an experimental study that assembles specialized automatic test generation tools designed to detect faults by estimating the likelihood of code being faulty. These tools will be compared against each other to discern their relative performance and effectiveness. Additionally, the study will comprehensively compare developer-generated tests with automatically generated tests to evaluate their respective aptitude for fault detection. Through this investigation, we seek to identify the most effective automated test generation tool while providing valuable insights into the relative merits of developer-generated and automatically generated tests for fault detection.