A company is developing a new model to predict the prices of specific items. The model performed well on the training dataset. When the company…

The scenario describes a classic case of overfitting, where the model performs exceptionally well on the data it was trained on but fails to generalize to new, unseen data in production. The most effective and fundamental way to combat overfitting is to train the model on a larger and more diverse dataset. By increasing the volume of training data, the model is exposed to more examples and variations, which helps it learn the true underlying patterns and relationships rather than memorizing the noise and specific quirks of a smaller dataset. This leads to better generalization and improved performance on new data.

1. Ng

A. (n.d.). CS229 Machine Learning Course Notes: Advice for Applying Machine Learning. Stanford University. In the section "Debugging: Bias vs. Variance

" getting more training examples is listed as the primary solution for a high variance (overfitting) problem.

2. Domingos

P. (2012). A few useful things to know about machine learning. Communications of the ACM

55(10)

78–87. On page 80

under the section "Overfitting: Your Worst Enemy

" the author states

"The simplest [defense against overfitting] is to get more data." DOI: https://doi.org/10.1145/2347736.2347755

3. Amazon Web Services. (2023). Amazon SageMaker Developer Guide: Prevent Overfitting. Under the section "Best practices to prevent overfitting

" the guide lists "Train with more data" as a key method

explaining that it helps the algorithm detect the relevant signal more easily.