MATLAB Deep Learning Accuracy Formula

Hello everyone, I want to classify a model with LSTM in MATLAB, but since the target values are between 1 and 5, 1 different predictions made by the model are taken as 0 accuracy. I want to use a new accuracy calculation formula based on the differences so that when it misclassifies with 1 difference (the difference is important because it is a scoring classification), the accuracy value will not be 0. How can I change the formula used when calculating training accuracy?

Thanks everyone in advance.

Answer 1

[aditi bagora]

Hello Burak,

I understand that you want to change the training accuracy formula. I would recommend defining a custom function for accuracy. I understand that if the difference between the predicted and actual class is less than or equal to 1 it should be considered as correct prediction otherwise it should be considered as misclassified.

Following approach can be taken to define the function:

• Get the predicted values from the LSTM model and the target class values.
• Initialize “correctPredictions = 0”.
• Compare each predicted value to target class value.
• If the difference between these values is less than or equal to 1.
• Increment “correctPredictions”.
• Compute accuracy by using “correctPredictions”.

Hope this helps!

Regards,

Aditi