Outliers Handling:
- Remove Outliers as many as possible. (Using smoothing method, moving average/ Savizy-Golay Method)
- Use the robust algorithms: Tree Algorithms (decision tree)/ Regularization(SVM, slack variable )
Regression base model is more senstive to outliers. - Use different metric (example: median instead of mean, absolute deviation instead of standard deviation): this method is mostly applied when outliers probably distort the distribution.
- some are resistant to outliers: tree, random forest, clustering, k-nearest neighbors.
