This paper presents HFS–Spearman, a label-agnostic hierarchical feature selection method for flow-based intrusion detection. The approach transforms Spearman rank correlations into a distance metric (d = 1 − |ρ|) and applies Ward-linkage agglomerative clustering to build a dendrogram that exposes dependency blocks among telemetry features. A compact representation is then constructed by selecting one exemplar (medoid) per block, yielding a non-redundant feature basis rather than a ranked list. The truncation level (λ*) is chosen adaptively using structure-preservation and partition-robustness diagnostics, avoiding fixed correlation thresholds. Experiments across heterogeneous datasets show consistent feature set compression (e.g., 46→23, 69→32) with reduced computational cost while maintai

Хажжуз А. (науч. рук. Авксентьева Е.Ю.) A hierarchical correlation-aware feature selection method for flow-based ids // Сборник тезисов докладов конгресса молодых ученых. Электронное издание. – СПб: Университет ИТМО, [2026]. URL: https://kmu.itmo.ru/digests/article/16329