MLS-C01 · Question #151
A Machine Learning Specialist is training a model using a supervised learning algorithm. The Specialist split the dataset to use 80% of the data for training and reserved 20% of the data for…
The correct answer is B. The mode is currently overfitting the training data and not performing as well as it should on data. A significant gap between high training accuracy and much lower test accuracy indicates that the machine learning model is overfitting. The model has learned the training data too specifically and fails to generalize to new, unseen data.
Question
A Machine Learning Specialist is training a model using a supervised learning algorithm. The Specialist split the dataset to use 80% of the data for training and reserved 20% of the data for testing. While evaluating the model the Specialist discovers that the mode is 97% accurate for the training dataset and 75% accurate for the test dataset. What is the reason for the discrepancy and what action should the Specialist take?
Options
- AThe high accuracy for the larger amount of training data means that the model is finished.
- BThe mode is currently overfitting the training data and not performing as well as it should on data
- CAdditional data in the test set is needed to balance the scoring of the model.
- DThe mode is currently underfitting and does not have enough complexity to capture the full scope
How the community answered
(29 responses)- A14% (4)
- B76% (22)
- C7% (2)
- D3% (1)
Why each option
A significant gap between high training accuracy and much lower test accuracy indicates that the machine learning model is overfitting. The model has learned the training data too specifically and fails to generalize to new, unseen data.
High training accuracy, especially when combined with low test accuracy, does not signify a finished model; instead, it indicates a problem with generalization.
The large discrepancy between 97% training accuracy and 75% test accuracy is a classic sign of overfitting, where the model has learned the training data too closely, including noise, and consequently performs poorly on new, unseen data. To address this, the specialist should implement regularization techniques, simplify the model, or consider early stopping during training.
The problem is not necessarily the quantity of test data but rather the model's inability to generalize from the training data, indicating overfitting rather than a need for more test data to 'balance scoring.'
Underfitting occurs when a model performs poorly on both training and test data due to insufficient complexity; the given scenario with high training accuracy contradicts this.
Concept tested: Overfitting and underfitting identification and mitigation
Source: https://developers.google.com/machine-learning/crash-course/generalization/peril-of-overfitting
Topics
Community Discussion
No community discussion yet for this question.