Now you are in the 1st Lithuanian (.LT) domain. 
26 October 2026, 13:00
Vilnius, Akademijos St. 4, Room 203
Online via MS Teams (https://bit.ly/DMSTI_2026-10-26)
dr. Dalia Breskuvienė
(Cognitive Computing Group)
„Heskes-Augmented Task-Aligned Feature Selection for Imbalanced Fraud Detection (HAT-FIDSOM) under Temporal Validation“
Annotation: Inspired by research conducted as part of a doctoral dissertation, HAT-FIDSOM addresses feature selection for fraud detection in highly imbalanced transaction data. The approach combines Heskes self-organizing maps with consensus ranking and task-specific relevance. This combination produces a significantly reduced set of features selected according to both the structure of the data and their relevance to fraud detection. Chronological validation reflects practical deployment, where models learn from past transactions to detect fraud in future transactions, without using future information during model selection.