Graph Theory · Community Detection · Visualization

Network
Analysis.

Every dataset hides a web of relationships. Network analysis makes those relationships visible — and actionable. I map connections, detect communities, identify bottlenecks, and surface insights that row-by-row analysis will never find.

Discuss a project Back to portfolio
Node ANode B Node CNode D Node ENode F HUB
The discipline

What is network analysis?

Network analysis studies the relationships between entities — not just the entities themselves. It answers questions that conventional analysis cannot reach.

// Graph Theory

Nodes and Edges

Every network is a graph: entities become nodes, relationships become edges. The mathematics of graphs reveals structure invisible in any spreadsheet.

// Community Detection

Clusters and Groups

Algorithms identify natural groupings within complex networks — revealing which entities belong together, and which connections bridge different communities.

// Centrality Metrics

Influence and Hubs

Degree, betweenness, eigenvector centrality — measuring which nodes matter most, which connections control information flow, and where vulnerabilities lie.

Case study

Criminal Network Study —
The D.S. Organisation

A real applied study tracking the structural evolution of a criminal trafficking network across 11 phases of a law enforcement investigation. Goal: identify key nodes, trace how the network reorganised after disruption, and determine the most effective intervention points.

The network consists of 110 individuals — players n1–n82 in trafficking roles, n83–n110 in support roles. The investigation initially targeted the leader, D.S. Wiretapped correspondence between 23 selected members was encoded into adjacency matrices, one per phase, and loaded into NetworkX graphs for temporal analysis.

Three centrality measures — Degree, Betweenness, and Eigenvector — were computed per player per phase. Mean centrality across all phases revealed who was structurally important throughout the investigation, and who rose or fell in influence after each disruption event.

Node & edge evolution — phases 1 to 11
Nodes Edges
Phase 1
15 · 18
Phase 2
24 · 28
Phase 3
33 · 56
Phase 4 seizure
33 · 48
Phase 5
32 · 39
Phase 6
27 · 47
Phase 7
36 · 49
Phase 8
42 · 58
Phase 9
34 · 44
Phase 10
42 · 50
Phase 11
41 · 50
ds_network_study.py
# Load each phase CSV into a NetworkX graph for idx, filename in enumerate(files): index = int(re.findall('[0-9]+', filename)[0]) phases[index] = pd.read_csv(filename, index_col=0) phases[index].columns = "n" + phases[index].columns G[index] = nx.from_pandas_adjacency(phases[index]) # Track centrality evolution across all 11 phases for phase in range(1, 12): deg[phase] = nx.degree_centrality(G[phase]) btw[phase] = nx.betweenness_centrality(G[phase]) eig[phase] = nx.eigenvector_centrality(G[phase]) # Mean centrality per player — who matters most overall? mean_deg = {n: np.mean( [deg[p][n] for p in deg if n in deg[p]] ) for n in all_nodes}
01

Phase 4 was the structural turning point

The first law enforcement seizure forced the organisation to abandon existing routes and pivot to cocaine imports from Colombia via the United States. Edge count dropped from 56 to 48 — broken connections made visible.

02

Node n12 emerged as the new hub in Phase 5

A new sub-network centred on n12 crystallised in Phase 5 to facilitate the new import operation. All three centrality metrics — degree, betweenness, eigenvector — confirmed n12's rapid rise in structural importance.

03

Centrality revealed the right targets, not just the most active

Temporal centrality tracking identified which players were structurally critical at each phase. This methodology applies directly to high-dimensional network security and any domain where disrupting a network — not just monitoring it — is the goal.

Structural analysis

Phase 4 vs Phase 11 —
The Network Transformed

The first seizure in Phase 4 was the pivotal event. Traffickers were forced to abandon existing routes and reorient cocaine imports from Colombia, transiting through New York. The two diagrams below show 23 selected members before and after this structural shift — and the emergence of a new hub.

n1 n2 n3 n4 n5 n6 n7 n8 n9 n10 n11 n12 n13 n14 n15 n16 n17 n18 n19 n20 n21 n22 n23 Phase 4 33 Nodes · 48 Edges · First Seizure

Node n1 acts as the primary hub. Node n12 has minimal connections — barely active in the network at this stage.

n1 n2 n3 n4 n5 n6 n7 n8 n9 n10 n11 n12 n13 n14 n15 n16 n17 n18 n19 n20 n21 n22 n23 Phase 11 41 Nodes · 50 Edges · n12 Dominant

Node n12 has emerged as a dominant hub, controlling the Colombia–New York cocaine import route and connecting multiple sub-networks.

The Phase 4 Event — What the Data Reveals

01

First Major Event: Phase 4 Seizure

The first law enforcement seizure happened in Phase 4. This forced the trafficking organisation to reorient entirely — pivoting to cocaine imports from Colombia, transiting through New York. The immediate structural effect is visible: edges dropped from 56 (Phase 3) to 48, representing broken connections as routes were abandoned and contacts cut.

02

Node n12 Emerges in Phase 5

The direct structural response to the Phase 4 seizure was the emergence of a new sub-network in Phase 5, centred on node n12 and its connections. This sub-network was built specifically to facilitate the new Colombia import operation. What was a peripheral node became the operational backbone of the reorganised network.

03

Centrality Measures Confirm the Shift

This inference is directly supported by the temporal evolution of all three centrality metrics. Across Degree, Betweenness, and Eigenvector centrality, n12 shows a measurable and statistically significant rise in influence from Phase 4 to Phase 5 — making it identifiable as the key intervention target in the reorganised network.

Exact Results — All 11 Phases

Phase Nodes Edges
Phase 1 15 18
Phase 2 24 28
Phase 3 33 56
Phase 4 seizure 33 48
Phase 5 32 39
Phase 6 27 47
Phase 7 36 49
Phase 8 42 58
Phase 9 34 44
Phase 10 42 50
Phase 11 41 50

The edge count drop in Phase 4 (56 → 48) and partial recovery in Phase 6 (47) reflect the network's disruption and subsequent reorganisation around the new Colombia route. Phase 8 marks the peak expansion: 42 nodes and 58 edges — the largest the network reached during the investigation.

Project delivery
4.85/5.0
Client satisfaction score
Project Delivery
Excellence

The investigators got their answers. The city got a little safer. The network got a little shorter. IntelXData got a 4.85 — which, statistically speaking, is as close to perfect as humans tend to agree on anything.

SCOPE
Dataset Scope Analyzed 22 weeks of wiretap correspondence data across 110 network members and 11 investigation phases.
TIME
Delivery Timeline Results fully prepared and demonstrated to the investigator group within 1 month of project initiation.
SCORE
Client Satisfaction Achieved a 4.85 / 5.0 rating from the investigator group — covering analytical rigor, clarity of findings, and delivery quality.
REACH
Broader Impact The methodology developed is applicable to high-dimensional network security, supply chain resilience, and organisational analysis.
My approach

How I work.

A structured process from raw relational data to actionable network intelligence.

01
Data Ingestion
Parsing relational data from SQL, APIs, CSVs, or SAP into adjacency structures.
02
Graph Construction
Building the network model — nodes, edges, weights, and directionality.
03
Metric Computation
Centrality, clustering coefficients, shortest paths, and community structure.
04
Visualization
Interactive diagrams via NetworkX, Plotly, and Tableau for any audience.
05
Insight Delivery
Clear narrative findings — in English, French, or German.
Tools used

Technical stack.

The instruments behind the analysis.

Python NetworkX Pandas NumPy SciPy Seaborn Matplotlib Plotly Tableau Gephi R (igraph) SQL Jupyter Notebooks
"Most analytical tools ask what happened.
Network analysis asks why it spread."
— Xavier Richert
Start a conversation All projects