Machine Learning and Neural Networks AIE Practice Questions (Page 3)

Questions 11–15

1. 11expert · hard

   A data science team is training a neural network for a binary classification task. The dataset is highly imbalanced: 95% of samples are class A and 5% are class B. The model currently achieves 95% accuracy but fails to identify most class B samples. Which change is most likely to improve the model's ability to detect class B?

   • AIncrease the number of hidden layers to make the model more complex
   • BSwitch to a regression model to predict a continuous score
   • CUse a loss function that penalizes misclassification of the minority class more heavily, such as weighted cross-entropy
   • DRemove the activation function from the output layer
   Reveal Solution Discussion Select an answer first
2. 12expert · hard

   A company is developing an AI system to control traffic lights to minimize average wait times. The system can change light timings based on real-time traffic conditions and receives a reward when wait times decrease. However, the system must also avoid causing gridlock, which is penalized. Which approach best balances learning to reduce wait times while avoiding gridlock?

   • AReinforcement learning with a reward function that includes both wait-time reduction and a penalty for gridlock
   • BUnsupervised clustering of traffic patterns to identify congestion hotspots
   • CSupervised learning on historical traffic data to predict wait times
   • DA fixed rule-based system that changes lights based on time of day
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A deep learning engineer is training a large neural network. The training loss decreases initially but then plateaus, and the validation loss starts to increase. The engineer suspects overfitting. Which combination of actions is most likely to reduce overfitting while maintaining model performance?

   • AAdd dropout layers and increase the amount of training data through augmentation
   • BDecrease the learning rate to zero to stop training
   • CIncrease the number of hidden layers and train for more epochs
   • DRemove the activation functions to make the model linear
   Reveal Solution Discussion Select an answer first
5. 14expert · medium

   A healthcare company wants to predict the likelihood of a patient developing a certain disease within five years. They have historical data with many features, including some missing values. The target is a binary label ('will develop' or 'will not develop'). They need to choose a model that can handle missing data and provide a probability estimate. Which approach is most appropriate?

   • AUse a reinforcement learning agent to learn a policy for disease prediction
   • BUse a supervised classification model like logistic regression after imputing missing values, which provides probability outputs
   • CUse a regression model to predict a continuous risk score and then threshold it
   • DUse an unsupervised clustering algorithm to group patients and assign risk based on cluster membership
   Reveal Solution Discussion Select an answer first
6. 15foundation · easy

   A data scientist has a dataset of customer records with no predefined categories. They want to discover natural groupings of customers based on purchasing behavior. Which type of machine learning is most appropriate?

   • ASemi-supervised learning
   • BUnsupervised learning
   • CReinforcement learning
   • DSupervised learning
   Reveal Solution Discussion Select an answer first