Performance Analysis

In the following section, we will first perform error analysis using the accuracy metric as a tool to determine the condition that is affecting the performance of the algorithm in greater proportion. Once the model is diagnosed, the hyperparameters can be tuned to improve the overall performance of the algorithm. The final model will be compared to those that were created during the previous chapter in order to determine whether a neural network outperforms the other models.

Error Analysis

Using the accuracy score calculated in Activity 14, we can calculate the error rates for each of the sets and compare them against each other to diagnose the condition that is affecting the model. To do so, a Bayes Error equal to 1% will be assumed, considering that other models in the previous chapter were able to achieve an accuracy level over 97%:

Figure 5.11: Accuracy score and error rate of the network

Note

Remember that in order to detect the condition that is affecting the network...