Machine Learning Model Improvement

Overview

There are generally three ways to try to improve a model.

• Collect More Data - data should be collected in an ongoing process. When new data is available, models should be tested on the new data and possibly include additional fitting.
• Use a Different Feature Set - many times the features used in model fitting have a large impact on the result. Sometimes useful features are missing, other times there are too many features and the model is overfit. (see feature extraction)
• Tweak the Model - tweaking the model can include adding additional parameters, and sometimes even changing the model type.