AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Reduce the amount of training data

  Why it's wrong here

  Reducing the amount of training data is the opposite of what is needed when a model overfits. With fewer examples, the model has even less evidence to distinguish genuine signal from noise, so its variance increases and the training-to-test performance gap normally becomes larger. Overfitting arises from high model capacity relative to the data available, and shrinking the dataset only makes that mismatch worse.

• ✓

  Apply regularization to the model

  Why this is correct

  Regularization directly targets overfitting by adding a penalty term to the loss function, such as an L1 (Lasso) or L2 (Ridge) weight penalty, or by using dropout in neural networks. This constrains the magnitude of learned coefficients, reducing the model's effective capacity and forcing it to represent smoother, more general patterns. As a result, the model is less likely to memorize idiosyncratic details of the training set and typically generalizes better on unseen house-price data.

• ✗

  Remove some features from the dataset

  Why it's wrong here

  Removing some features can occasionally help if those features are pure noise or highly redundant, but it is a blunt and heuristic approach that risks discarding predictive information. Unlike regularization, which systematically shrinks the influence of all features while retaining them, feature deletion does not adapt to the strength of each feature's relationship with the target. In many overfitting scenarios, the problem is not the presence of irrelevant inputs but the model's freedom to fit complex interactions, so deletion often provides little benefit.

• ✗

  Increase the number of layers in the neural network

  Why it's wrong here

  Increasing the number of layers in the neural network raises its parameter count and expressiveness, which typically makes overfitting more severe rather than less. If the model is already memorizing training data, adding more layers gives it even greater ability to fit noise and spurious patterns, widening the gap between training and validation performance. The appropriate response to overfitting is to reduce capacity or add constraints, not to increase architectural complexity.

Go deeper

Related to this question