Detecting outliers using LOF
In the previous recipe, Detecting outliers using KNN, in the KNN algorithm, the decision scoring for detecting outliers was based on the distance between observations. A data point far from its KNN can be considered an outlier. Overall, the algorithm does a good job of capturing global outliers, but those far from the surrounding points may not do well with identifying local outliers.
This is where the LOF (Local Outlier Factor) comes in to solve this limitation. Instead of using the distance between neighboring points, it uses density as a basis for scoring data points and detecting outliers. The LOF is considered a density-based algorithm. The idea behind the LOF is that outliers will be further from other data points and more isolated, and thus will be in low-density regions.
It is easier to illustrate this with an example: imagine a person standing in line in a small but busy Starbucks, and everyone is pretty much close to each other; then, we can say...
