<aside> 💡 Source Code for the project here: https://github.com/AstuteSource/SEERS

</aside>

GitHub - AstuteSource/SEERS: This repository serves as the public workspace for the software engineering cohort of the CIS department's 2024 Junior Seminar course.

Challenge: Mutatino testing, a valuable software testing technique, can be computationally expensive. Companies often spend significant time and resources running mutation tests on their code

Task: In collaboration with another student and Dr.Gregory M. Kapfhammer, I aimed to develop a tool that could predict mutation scores earlier in the testing process. This would allow developers to focus their efforts on mutations more likely to reveal defects and improve overall testing efficiency

Action: We focus on Python projects and investigated the use of machine learning models to predict mutation scores based on the analysis of code anti-patterns.

• Code Analysis: We leveraged the Chasten tool to identify code anti-patterns and Mutmut to generate the actual mutations
• Data Preparation: the information from both tools was merged into a unified dataframe
• Machine Learning Exploration: I experimented with various machine learning algorithms to identify the most effective model for predicting mutation scores. Additionally, I explored cross-prediction to see if the model could generalize to projects beyond the training data.

DETAILED REPORT

1. Data Collection and Preprocessing

The dataset for study comprised 323 functions extracted from 12 open-source Python projects. Each function was analyzed to extract relevant information, including its code, mutation source, and a list of identified code patterns.

Data Preprocessing:

Data Preprocessing

1. Pattern Extraction and Refinement

The initial set of code patterns exhibited several issues: