What is model performance in machine learning?
Model performance in machine learning is a measurement of how accurate predictions or classifications a model makes on new, unseen data. You typically measure model performance using a test set, where you compare the predictions on the test set to the actual outcomes. For classification, you can use accuracy as a measure of model performance, but you cannot use accuracy as a measure of model performance for regression problems, where metrics such as mean absolute error and (root) mean squared error are used.
Causes of poor model performance
There can be many reasons for poor model performance, but some of the key things you should check for are:
- Overfitting - training for too long, so your model cannot generalize to perform well on unseen data;
- Data leakage - where the training data is polluted with test data or future data;
- Model complexity - has the model too many parameters compared to the number of training samples (a rule-of-thumb is that you should have 1 training sample for each model parameter/weight for DNNs);
- Data quality - is the data quality high enough?