R Deep Learning Essentials by Mark Hodnett

Author:Mark Hodnett
Language: eng
Format: epub
Tags: COM042000 - COMPUTERS / Natural Language Processing, COM004000 - COMPUTERS / Intelligence (AI) and Semantics, COM044000 - COMPUTERS / Neural Networks
Publisher: Packt Publishing
Published: 2018-08-24T10:31:09+00:00

Figure 6.1: An example of a learning curve which plots accuracy by data size

In this case, accuracy is in a very narrow range and stabilizes as the # instances increase. Therefore, for this algorithm and hyperparameter choice, adding more data will not increase accuracy significantly.

If we get a learning curve that is flat like in this example, then adding more data to the existing model will not increase accuracy. We could try to improve our performance by either changing the model architecture or by adding more features. We discussed some options for this in Chapter 5, Image Classification Using Convolutional Neural Networks.

Going back to our binary classification model, let's consider how we could we use it in production. Recall that this model is trying to predict if customers will return in the next x days. Here is the confusion matrix from that model again:

Predicted

Actual 0 1

0 10714 4756

1 3870 19649

If we look at how the model performs for each class, we get a different accuracy rates:

For Actual=0, we get 10714 / (10714 + 4756) = 69.3% values correct. This is called specificity or the true negative rate.

For Actual=1, we get 19649 / (3466 + 19649) = 85.0% values correct. This is called sensitivity or the true positive rate.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.