TLDR Tech / InfoSec
Threat Hunting Using Pair Probabilities
A KQL threat hunting technique built on Microsoft's pair_probabilities_fl() user-defined function calculates how statistically unusual a categorical pair (like a binary and its execution directory) is across a dataset, surfacing outliers such as cmd.exe launching from a user's Desktop instead of System32. The author extended the function to return top co-occurring values, per-device context, and occurrence counts alongside the core probability metrics (P_A|B, Lift, and Jaccard), making rare pairings easier to triage. The approach is explicitly framed as a statistical rarity detector rather than a maliciousness classifier, useful for narrowing investigation scope but unreliable on small datasets or environments where malicious activity has already been baked into the baseline.
Read the full story