# The impact of feature importance methods on the interpretation of defect classifiers

arXiv: https://arxiv.org/abs/2202.02389 · HTML: https://mlsanity.com/paper/2202.02389/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 138 references that are in the map.
- First posted on arXiv 4 Feb 2022.

## Builds on (most cited first, 21 of 21)

- [A Unified Approach to Interpreting Model Predictions](https://mlsanity.com/paper/1705.07874/) (2017)
- ["Why Should I Trust You?": Explaining the Predictions of Any Classifier](https://mlsanity.com/paper/1602.04938/) (2016)
- [The Mythos of Model Interpretability](https://mlsanity.com/paper/1606.03490/) (2016)
- [A Benchmark for Interpretability Methods in Deep Neural Networks](https://mlsanity.com/paper/1806.10758/) (2018)
- [All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously](https://mlsanity.com/paper/1801.01489/) (2018)
- [Consistent Individualized Feature Attribution for Tree Ensembles](https://mlsanity.com/paper/1802.03888/) (2018)
- [The Impact of Automated Parameter Optimization on Defect Prediction Models](https://mlsanity.com/paper/1801.10270/) (2018)
- [The Impact of Class Rebalancing Techniques on the Performance and Interpretation of Defect Prediction Models](https://mlsanity.com/paper/1801.10269/) (2018)
- [Improving Palliative Care with Deep Learning](https://mlsanity.com/paper/1711.06402/) (2017)
- [Tuning for Software Analytics: is it Really Necessary?](https://mlsanity.com/paper/1609.01759/) (2016)
- [Explainable Software Analytics](https://mlsanity.com/paper/1802.00603/) (2018)
- [Towards Understanding and Detecting Fake Reviews in App Stores](https://mlsanity.com/paper/1904.12607/) (2019)
- [Applications of Psychological Science for Actionable Analytics](https://mlsanity.com/paper/1803.05067/) (2018)
- [Disentangled Attribution Curves for Interpreting Random Forests and Boosted Trees](https://mlsanity.com/paper/1905.07631/) (2019)
- [ConPredictor: Concurrency Defect Prediction in Real-World Applications](https://mlsanity.com/paper/1804.03589/) (2018)
- [AutoSpearman: Automatically Mitigating Correlated Metrics for Interpreting Defect Models](https://mlsanity.com/paper/1806.09791/) (2018)
- [Predicting Good Configurations for GitHub and Stack Overflow Topic Models](https://mlsanity.com/paper/1804.04749/) (2018)
- [An empirical assessment of best-answer prediction models in technical Q&A sites](https://mlsanity.com/paper/1903.09522/) (2019)
- [Is the Stack Distance Between Test Case and Method Correlated With Test Effectiveness?](https://mlsanity.com/paper/1903.05432/) (2019)
- [How to Improve AI Tools (by Adding in SE Knowledge): Experiments with the TimeLIME Defect Reduction Tool](https://mlsanity.com/paper/2003.06887/) (2020)
- [Does chronology matter in JIT defect prediction? A Partial Replication Study](https://mlsanity.com/paper/2103.03506/) (2021)

## Built on by (most cited first, 12 of 12)

- [VulScribeR: Exploring RAG-based Vulnerability Augmentation with LLMs](https://mlsanity.com/paper/2408.04125/) (2024)
- [What Were You Thinking? An LLM-Driven Large-Scale Study of Refactoring Motivations in Open-Source Projects](https://mlsanity.com/paper/2509.07763/) (2025)
- [Predicting the First Response Latency of Maintainers and Contributors in Pull Requests](https://mlsanity.com/paper/2311.07786/) (2023)
- [McUDI: Model-Centric Unsupervised Degradation Indicator for Failure Prediction AIOps Solutions](https://mlsanity.com/paper/2401.14093/) (2024)
- [CID: Measuring Feature Importance Through Counterfactual Distributions](https://mlsanity.com/paper/2511.15371/) (2025)
- [SimClone: Detecting Tabular Data Clones using Value Similarity](https://mlsanity.com/paper/2407.12802/) (2024)
- [Feature Importance Depends on Properties of the Data: Towards Choosing the Correct Explanations for Your Data and Decision Trees based Models](https://mlsanity.com/paper/2502.07153/) (2025)
- [A Study on Mixup-Inspired Augmentation Methods for Software Vulnerability Detection](https://mlsanity.com/paper/2504.15632/) (2025)
- [On The Impact of Merge Request Deviations on Code Review Practices](https://mlsanity.com/paper/2506.08860/) (2025)
- [Dynamic Features Adaptation in Networking: Toward Flexible training and Explainable inference](https://mlsanity.com/paper/2510.08303/) (2025)
- [CAFE-GB: Scalable and Stable Feature Selection for Malware Detection via Chunk-wise Aggregated Gradient Boosting](https://mlsanity.com/paper/2601.15754/) (2026)
- [Cross-Course Generalizability of SRL-Aligned Predictive Models Using Digital Learning Traces](https://mlsanity.com/paper/2604.22812/) (2026)

## Abstract (arXiv)

Classifier specific (CS) and classifier agnostic (CA) feature importance methods are widely used (often interchangeably) by prior studies to derive feature importance ranks from a defect classifier. However, different feature importance methods are likely to compute different feature importance ranks even for the same dataset and classifier. Hence such interchangeable use of feature importance methods can lead to conclusion instabilities unless there is a strong agreement among different methods. Therefore, in this paper, we evaluate the agreement between the feature importance ranks associated with the studied classifiers through a case study of 18 software projects and six commonly used classifiers. We find that: 1) The computed feature importance ranks by CA and CS methods do not always strongly agree with each other. 2) The computed feature importance ranks by the studied CA methods exhibit a strong agreement including the features reported at top-1 and top-3 ranks for a given dataset and classifier, while even the commonly used CS methods yield vastly different feature importance ranks. Such findings raise concerns about the stability of conclusions across replicated studies. We further observe that the commonly used defect datasets are rife with feature interactions and these feature interactions impact the computed feature importance ranks of the CS methods (not the CA methods). We demonstrate that removing these feature interactions, even with simple methods like CFS improves agreement between the computed feature importance ranks of CA and CS methods. In light of our findings, we provide guidelines for stakeholders and practitioners when performing model interpretation and directions for future research, e.g., future research is needed to investigate the impact of advanced feature interaction removal methods on computed feature importance ranks of different CS methods.