Chapter 5 – Improving the Performance of Machine Learning Models
- Adding more training data points could help to reduce variance while adding more features could help to reduce bias. However, there is no guarantee of a reduction of variance through the addition of new data points or whether new features will be helpful in reducing variance.
Assigning weights during optimization: You can assign a weight to each data point, according to the confidence of class labels, when training machine learning models.
Ensemble learning: If you consider a distribution of the quality or confidence score of each data point, then you can build different models using data points from each part of this distribution and then combine the predictions of the models for example using their weighted average.
Transfer learning: You can train a model on a large dataset with different levels of label confidence (see Figure 5.3), excluding very low-confidence data and then fine-tune it on...
