Your ML model is making poor predictions, and score on test data is poor. What to do?
Tracking and diagnosing what the issue is would be a good starting point.
Enter Learning Curves.
Learning curves show how has your model learnt or performed on the train and validation data.
While cross validation, we can plot 2 curves - one for loss on training data and other on the loss on validation data.
Now, if your training loss just doesn’t decrease with more data points, showing a flat line, it means that your model is Underfitting.
It just isn’t able to learn the data.
Or there could also be a case when both the training and validation loss keep decreasing and show signs of a further reduction when the training is complete.
However, when the learning curve for training loss decreases to a certain point and gets stable, but the learning curve for validation loss starts to increase after a certain point, it is a sign of Overfitting.
And if the learning curves for both training loss and validation loss look the same, that is, decrease to a certain point and get stable, that means your model is having a Good fit.
Learning curves are a nice way to track your model. Have you used them yet?
#datascience #machinelearning