The idea
This research explored how transparent machine-learning systems could support software engineering decisions. The work paired a defect-prediction model with a structured responsible-AI framework and explanations intended to improve user understanding.
Approach
We co-developed a four-phase framework and evaluated it using NASA JM1 defect-prediction data. SHAP-based explanations were included to help users understand model outputs. The source material reports approximately 82% AUC and a 15–20% improvement in user comprehension.
Reported outcomes
Tools & methods
More about the project
The project focused on responsible model use in software engineering decision support. It combines model evaluation with explanation and reported user-comprehension outcomes.
What I took from it
Responsible-AI work needs to address how people interpret system outputs as well as how a model performs on a dataset.