Team Big D | LSESU Datathon 2026

London Underground:
Siting the Next Station

Geospatial Analysis

LSESU Data Science Datathon 2026

Rex, Ling Ya, Chang, Vignesh, Sid

The Challenge

Choose the best place to implement a new tube station.

1. Managing Equity

Addressing deprivation and transit poverty in underserved communities across Greater London.

2. Managing Demand

Population density, bus crowding, employment hubs — where do people actually need to travel?

3. Supply

Network capacity and infrastructure constraints — can the existing grid handle a new node?

How do we mathematically solve for all three?

Our Approach

1
Spatial Data
PTAL, IMD &
Demographics
▸
2
Demand Scoring
Weighted Need
Score → Top 10
▸
3
Network Graph
NetworkX
Topology
▸
4
Stress Test
Station Insertion
Simulation
▸
5
MCDA
Multi-Criteria
Decision Analysis
▸
6
Robustness
Monte Carlo
10,000 sims

A 6-stage pipeline from raw spatial data to a statistically validated recommendation

Identifying Transit Deserts

Demand Modeling

We computed a weighted "Need Score" combining 6 normalised indicators:

  • Inverse PTAL (Transit Poverty)
  • Distance to nearest station
  • Population density
  • Bus crowding data
  • Employment density
  • IMD (Deprivation)
$\text{Need}_i = \sum_{k} w_k \cdot \hat{x}_{ik} \quad \text{(min-max scaled)}$

The Shortlist

Filtered 4,994 LSOAs to
Top 10
Geographically spaced (max 2/borough, > 2 km apart)

Output: Top 10 candidates exported as CSV with Need Scores, coordinates, and catchment stats.

The Network Topology

We reconstructed the London Underground as a mathematical graph using NetworkX — a Python library for creating, manipulating, and studying complex networks.

Each station = node; each connection = weighted edge. This lets us compute shortest paths, global efficiency, and centrality across the network.

$C_B(v) = \sum_{s \neq v \neq t} \frac{\sigma(s,t \mid v)}{\sigma(s,t)}$

$\sigma(s,t)$ = shortest paths between s,t; $\sigma(s,t|v)$ = those through v. High $C_B$ = bottleneck.

358
Nodes (Stations)
492
Edges (Rail & Transfer)
$C_B(v)$
Betweenness Centrality — bottleneck stations controlling shortest-path traffic

Baseline Global Efficiency and shortest-path matrices established.

Stress-Testing the Network

We programmatically inserted each top-10 candidate into the mathematical graph to simulate the impact.

  • Global Efficiency Δ: % improvement in shortest paths across all node pairs.
  • Bottleneck Relief: Reduction in maximum betweenness centrality.
  • Improved OD Pairs: Origin-Destination pairs gaining faster routes.
  • Total Path Savings: Sum of kilometres saved city-wide.

Key Insight

We match high-need LSOA requirements with the existing network — finding the intersection of demand and topology.

Adding a new line segment actually increases overall travel time, as trains stop at every station along the route.

This led us to evaluate an interchange — connecting to existing lines rather than extending them.

The MCDA Decision Framework

Multi-Criteria Decision Analysis — all outputs scaled to a 0–1 index, weights allocated into 3 thematic tiers:

Local Demand & Equity (45%)
Network Flow & Relief (40%)
Feasibility (15%)
  • • Need Score (20%)
  • • Catchment Population (15%)
  • • IMD Deprivation (10%)
  • • Global Efficiency Δ (12%)
  • • Bottleneck Relief (10%)
  • • Improved OD Pairs (10%)
  • • Path Reduction (8%)
  • • Network Connections (7%)
  • • Path Savings km (8%)

Our Choice

Building Olympic Way Interchange

LSOA: Brent 035C

  • Ranked 1st across the MCDA matrix
  • Catchment population of 7,400+ residents within 800m
  • Bottom 20% in existing PTAL rankings (High Transit Poverty)
  • Topological improvements to Jubilee & Metropolitan lines as an interchange

Why an Interchange?

Maximises socio-economic ROI while acting as a pressure release valve for North-West London rail flows — without the travel-time penalty of extending a line.

Proving the Robustness

We ran 10,000 Monte Carlo simulations perturbing all MCDA weights by ±30% to ensure our results are structurally sound.

Olympic Way Win Rate
63%
Ranked 1st in 6,300 configurations

Key Findings

  • Clear Dominance: Next best alternative won only 14% of scenarios.
  • 95% Confidence Interval: ranked between 1.0 and 2.8 across perturbations.
  • Structural Resilience: Remains optimal even if 'Feasibility' weight is discounted entirely.

Limitations & Next Steps

Current Limitations

  • Existing Connectivity: Central London is already efficiently connected, limiting where new stations can have maximal impact.
  • Planned Expansions: The model does not incorporate data on future stations TfL may already be planning.
  • Subsurface Topology: Node mapping is an approximation; we lacked exact physical GPS routing data for existing underground lines.

Future Enhancements

  • Algorithm Upgrades: Implement more rigorous, weighted Dijkstra pathfinding.
  • Multi-Modal Data: Continue improving multi-modal data integration.
  • Overground Integration: Integrate deeply with Overground, DLR, and Great Western Rail to better evaluate areas like Peckham and Camberwell.