@@ -69,6 +69,8 @@ A simple way of doing so would be to apply a local KDE function and then apply a
However, the problem with the above approach is that one needs geometrical knowledge in order to define if a region in the KDE plot is actually an anomaly.
For instance, one might calculate the area of the regions, but then one needs to decide when the area should be considered as an anomaly.
But there is also an alternative: One could apply the mapper algorithm with the same filter function as seen in the "circles" demonstration:
...
...
@@ -79,6 +81,7 @@ Then one needs to scale the graph, and subsequently one can perform a nearest ne
The holes in the graph are now the topological aspect which represents the anomalies.
One can now develop an algorithm on the graph in order to find the cycles which encompass the anomalies.
The important aspect is that the anomalies in the graph are geometry independent, one only needs to find a good algorithm in order to find the cycles in the graph.
