Isolation Forest is an unsupervised, tree-based anomaly detection methodology. See how each KernelSHAP and TreeSHAP can be utilized to clarify its output.
Isolation Forest has turn out to be a staple in anomaly detection programs [1]. Its benefit is having the ability to discover advanced anomalies in giant datasets with many options. Nevertheless, in terms of explaining these anomalies, this benefit rapidly turns into a weak point.
To take motion on an anomaly we frequently have to know the explanations for it being categorised as one. This perception is especially invaluable in real-world purposes, similar to fraud detection, the place figuring out the rationale behind an anomaly is commonly as vital as detecting it.
Sadly, with Isolation Forest, these explanations are hidden inside the advanced mannequin construction. To uncover them, we flip to SHAP.
We’ll apply SHAP to IsolationForest and interpret its output. We’ll see that though that is an unsupervised mannequin we will nonetheless use SHAP to clarify its anomaly scores. That’s to know:
- How options have contributed to the scores of particular person cases
- and which options are vital usually.