2WIE 2026 Keynote

Reconstructing spatio-temporal criminal networks for disruption in the Amazon — mathematical-audience version

Contents

Abstract

Interceptions reveal locations and sequences, but not every actor or connection. The talk frames the problem as inference on the unobserved vertices and edges of a spatio-temporal criminal network, and proposes a reproducible build–reconstruct–validate pipeline:

  1. Chronnet construction — 586 HF communications from 2018 → 333 vertices and 1,105 directed edges, with accounting identities for verification.
  2. Supervised reconstruction — classification under extreme imbalance (≈ 1 % edges): PCA + SMOTE + logistic regression, selected by specificity (47 false positives vs. 334 and 1,339).
  3. Unobserved elements — 107 candidate unknown edges and 84 unique hidden vertices, with spatial validation (Wilcoxon tests by infrastructure and scale) and operational validation with CENAC-DIJIN and DIRAN.

Authors: Andrés Pérez-Coronado · Gustavo Canavire-Bacarreza · Santiago Ortíz · Henry Laniado (FastAnalytics S.A.S)

Scientific talk — 2nd International Statistics Workshop 2026, Universidad Nacional de Colombia, Manizales Campus · November 5–6, 2026 · La Nubia Campus, R and W blocks.

Slides

📥 Download the PDF (43 slides, 16:9)

The presentation is organized in three parts:

Part Contents
Problem and data Observable subnetwork, element typology (known, unknown, hidden), 586 HF records from 2018
Method Chronnet construction (equations 1–3), dyadic dataset, SMOTE and model selection, hidden-vertex search with homophily
Results and decision In-sample reconstruction (97 % of edges), unknown edges (107) and hidden vertices (84), spatial proximity rule, policy implications

Deck sources (Quarto/beamer, FastAnalytics style): presentaciones/matematicas/ in this repository.

About the event

The 2nd International Statistics Workshop 2026 — Methods, Modeling and Applications — takes place on November 5–6, 2026 at the La Nubia Campus (R and W blocks), Universidad Nacional de Colombia, Manizales, with free admission upon registration.