A company is training an Al model using a custom table to determine the amount of time it takes to deliver a package based on several key fields. The testing data used to train the model is used for…

Question

Options

• AReduce the size of the data used within the model.
• BIncrease the size of the data used with the model.
• CUse sample training data from Microsoft
• DReplace the training data with real-world data.

• A
  6% (2)
• B
  3% (1)
• C
  13% (4)
• D
  77% (24)

Explanation

The large gap between training performance (2% variance) and real-world performance (15% variance) is a classic sign that the training data does not represent actual conditions - a form of overfitting or data leakage caused by using the same dataset for all training and regression testing. The model learned patterns specific to that dataset rather than generalizable real-world patterns. The correct fix is (D) replace the training data with real-world data, which introduces genuine variability and will produce a model that generalizes properly. Option A (reducing data size) would worsen the problem. Option B (increasing data size) only helps if the data itself is representative - more of the same flawed data won't close the gap. Option C (Microsoft sample data) is unrelated to a package-delivery domain model.

No community discussion yet for this question.