Skip to content

Latest commit

 

History

History
21 lines (17 loc) · 925 Bytes

README.md

File metadata and controls

21 lines (17 loc) · 925 Bytes

ML_FaultUnderstanding

Goal:

Investigate the factors that impact the accuracy of fault understanding (ability to recognize the code that is causing a software failure). The analyzed factors consist of attributes of programmers (profession, year of experience, coding ability) and tasks (duration, confidence, difficulty).

Metrics:

  • Quit rate
  • Task uptake rate
  • Qualification score
  • Task duration
  • Explanation size
  • Inter-rater reliability

Data:

Two experiments on identifying the code that is causing a software failure. Experiments:

  • E1: 5405 tasks, 777 programmers, 10 real failures from 10 popular open source projects
  • E2: 2580 tasks, 497 programmers, 8 real failures from 5 popular open source projects

Analysis methods:

Experiments replication, descriptive statistics (ANOVA, Chi2square, Wilcoxon), correlation Analysis (Kendall-tau), predictive models (logistic regression,Random Forest).