Why are evaluation metrics not used as loss functions for the learning algorithm then?
When building a learning algorithm we are looking to maximize a given evaluation metric (say accuracy), but the algorithm will try to optimize a different loss function during learning (say MSE/entropy). Why are the evaluation metrics not used as loss functions for the learning algorithm then?
What do you need to know about loss functions?
What are loss functions? Loss functions (also known as objective functions) are equations that give you a curve of loss generated by the predictions of your model. Our aim is to minimize the loss function to enhance the accuracy of the model for better predictions.
When to use a multiclass classification loss function?
Multiclass classifications loss functions would be used for a problem involving more than two classes. For e.g. if we are trying to classify cuisines based on different ingredients required for the preparation, types of cuisines would become the class labels.
How are loss functions used in machine learning?
The loss functions should be able to penalize wrong prediction which is done with the help of negative log. If the prediction is correct and the probability is 1, the value will be zero which is ideal, but if it is lesser than zero then the number becomes negative.
When does loss error decrease your metric scores improve?
In general, when loss error decreases your metric scores improve. Therefore, the two are linked and sharing the same objective. It is a good question. I would like to answer it from a different perspective.
How are evaluation metrics used for classification problems?
For classification problems, metrics involve comparing the expected class label to the predicted class label or interpreting the predicted probabilities for the class labels for the problem. Selecting a model, and even the data preparation methods together are a search problem that is guided by the evaluation metric.
Can you choose a model based on loss error?
Choosing the best model based on loss error would not always work since they are not directly comparable. Therefore, metrics are used to evaluate your trained models. In general, when loss error decreases your metric scores improve. Therefore, the two are linked and sharing the same objective.