Hands-On Exercise: Build a Defect Prediction System CT-AI Practice Questions (Page 6)

Questions 26–30

1. 26foundation · easy

   What is the goal of feature selection in a defect prediction model?

   • ATo reduce the number of features to only those that are most relevant for predicting defects
   • BTo normalize the values of all features
   • CTo increase the number of features to improve model accuracy
   • DTo create new features by combining existing ones
   Reveal Solution Discussion Select an answer first
2. 27foundation · easy

   Why is data balancing important when preparing a defect prediction dataset?

   • ATo ensure the model is not biased toward the majority class, which is often the non-defective modules
   • BTo convert categorical data into numerical data
   • CTo increase the size of the dataset so that the model trains faster
   • DTo remove duplicate records from the dataset
   Reveal Solution Discussion Select an answer first
4. 28expert · hard

   A team is building a defect prediction model and has a large number of features. They want to reduce the feature set to improve model performance and interpretability. They have tried filter methods (e.g., correlation) and wrapper methods (e.g., recursive feature elimination). The wrapper method is computationally expensive. What is the most appropriate next step?

   • AUse all features and rely on the model's regularization to handle irrelevant ones.
   • BUse an embedded method like Lasso regression, which performs feature selection during model training.
   • CManually select features based on domain knowledge without any data-driven analysis.
   • DIncrease the computational resources to run the wrapper method.
   Reveal Solution Discussion Select an answer first
5. 29application · medium

   A team has collected defect data from multiple sources, including version control, issue tracker, and code review tools. They notice that some modules have missing values for certain metrics, and the data formats are inconsistent. What is the first step they should take to prepare the data for modeling?

   • ASelect features before cleaning the data.
   • BNormalize the data before cleaning it.
   • CClean the data by handling missing values and standardizing formats.
   • DTrain a model immediately to see what happens.
   Reveal Solution Discussion Select an answer first
6. 30application · medium

   A team is building a defect prediction model and has generated over 200 candidate metrics, including code churn, complexity, and developer experience. They are concerned about overfitting and want to improve model interpretability. Which approach is most appropriate?

   • AInclude all 200 metrics to ensure no information is lost.
   • BUse feature selection to identify the most relevant subset of metrics based on correlation with defect-proneness.
   • CRandomly select 20 metrics without looking at their relevance.
   • DUse only the newest metrics, assuming they are more relevant.
   Reveal Solution Discussion Select an answer first