Exam Professional Data Engineer topic 1 question 191 discussion

There 2 parts and they are relevant to each other 1. Overfit is fixed by decreasing the number of input features (select only essential features) 2. Accuracy is improved by increasing the amount of training data examples.

Option B, the model learned to listen to too much stuff/noise. We need to reduce it, by decreasing the number of input feature, and we need to give the model more data, by increasing the amount of training data

Increase the size of the training dataset: By adding more diverse examples of customers and their buying behavior to the training data, the model will have a broader understanding of different scenarios and be better equipped to generalize to new customers. Increase the number of input features: Providing the model with more relevant information about customers can help it identify meaningful patterns and make better predictions. These input features could include things like the customer's age, past purchase history, browsing behavior, or any other relevant data that might impact their buying likelihood.

A. can be a solution for a specific case, but it is not the academic answer as we do not know the quantity and proportion between them of n and k added. More records and more variables together can lead to even more overfitting due also to the curse of dimensionality. Adding a variable is much more impactful than records. B. just more records can lead to a more robust estimation and fewer variables certainly lead to at most the same estimation, but potentially reduce the fit on the training set. C. reduce n in favor of k is never a choice. it is against logic and it will lead to more overfitting. D. decrease both will reduce overfitting for sure but at the price of losing robustness on the model predictive power

B. Increase the size of the training dataset, and decrease the number of input features.

B is correct

Answer B https://machinelearningmastery.com/impact-of-dataset-size-on-deep-learning-model-skill-and-performance-estimates/

Option B Feature selection is the one the ways to resolve overfitting. Which means reducing the features when the size of the training data is small, then the network tends to have greater control over the training data. so increasing the size of data would help.

Best option is not mentioned: generalize you neural net by decreasing the complexity of it's structure. A part from that I guess you could remove some features and increase the size of the training dataset ==> B

B. Increase the size of the training dataset, and decrease the number of input features. Sorry, B is right. Read through extensive best-practices on ML.

D is correct

D. Reduce the size of the training dataset, and decrease the number of input features. Reveal Solution

B. Increase the size of the training dataset, and decrease the number of input features.