AI Life Cycle (Model Selection, Training, and Validation) AAISM Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   Which step is a best practice in the data preparation phase of model training?

   • ASplitting the data into training and test sets before any preprocessing to avoid data leakage
   • BNormalizing features after splitting the data using statistics from the entire dataset
   • CUsing the entire dataset for training to maximize model accuracy
   • DRemoving all missing values without considering the reason for missingness
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   When selecting an AI model for a fraud detection system, which criterion is most important to consider alongside predictive performance?

   • AThe model's default hyperparameter values
   • BThe model's training time on the available hardware
   • CThe model's popularity in the open-source community
   • DThe model's ability to provide explanations for its decisions
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A team is training a random forest model for a binary classification task. The training accuracy is 99%, but the test accuracy is 75%. The team has already tried increasing the number of trees and decreasing the maximum depth. Which additional strategy is most likely to reduce overfitting?

   • AIncrease the number of features considered at each split.
   • BReduce the number of trees in the forest.
   • CUse bootstrap sampling without replacement to create more diverse trees.
   • DIncrease the minimum samples per leaf to force the model to learn more general patterns.
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A team is developing a model to predict customer churn. They have a dataset of 10,000 customers, with 500 churners. They need to evaluate the model's ability to identify churners accurately. Which validation approach would provide the most reliable estimate of performance?

   • AUse a simple random split and report AUC-ROC.
   • BUse stratified k-fold cross-validation and report the F1-score.
   • CUse a single random 70/30 train-test split and report accuracy.
   • DUse holdout validation with a 50/50 split and report precision.
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   An organization needs to classify emails as spam or not spam. Which model selection criterion is most directly relevant when choosing between a logistic regression and a deep neural network for this task?

   • AThe problem type (binary classification) and the size and quality of the labeled dataset
   • BThe color scheme of the model's visualization dashboard
   • CThe interpretability of the model for stakeholders
   • DThe geographic location of the data center hosting the model
   Reveal Solution Discussion Select an answer first