A health care company is planning to use neural networks to classify their X-ray images into normal and abnormal classes. The labeled data is divided into a training set of 1,000 images and a test…

Question

Options

• AChoose a higher number of layers
• BChoose a lower number of layers
• CChoose a smaller learning rate
• DEnable dropout
• EInclude all the images from the test set in the training set
• FEnable early stopping

• A
  9% (3)
• B
  73% (24)
• C
  15% (5)
• E
  3% (1)

Explanation

The model has 99% training accuracy but only 55% test accuracy - a classic sign of severe overfitting. With only 1,200 images and 50 hidden layers, the model is massively over-parameterized. The correct fixes are: (B) Reduce the number of layers - fewer layers means a simpler model with less capacity to memorize training data, reducing overfitting. (D) Enable dropout - dropout randomly deactivates neurons during training, acting as a regularizer that forces the network to learn redundant representations and generalize better. (F) Enable early stopping - monitoring validation loss and halting training when it begins to increase prevents the model from over-training on the training set. Options to avoid: increasing layers (A) worsens overfitting; a smaller learning rate (C) slows convergence but does not fix overfitting; including test data in training (E) is data leakage and invalidates evaluation.

No community discussion yet for this question.