Home/Soccer Analytics/Lineup Optimization via Mixed Integer Programming

Lineup Optimization via Mixed Integer Programming

Tactical AnalysisLevel 4 — Expert

Prerequisites

Player Pair Synergy Scoring

What It Is

Using mixed integer programming (MIP) to select the optimal 11 players from a squad of ~25 that maximizes total pair synergy while meeting formation constraints (1 GK, N defenders, N midfielders, N forwards). With 25 players, there are ~4.5 million possible lineup combinations. MIP solves this in seconds by formulating team selection as a constrained optimization problem rather than brute-force enumeration. The key advancement over individual-maximizing lineups: the pair-based MIP considers how selected players interact with each other, not just their solo values.

Correct Execution

(1) Individual-only MIP (baseline): Maximize Σ V(player_i) × X_i, where X_i is a binary decision variable (1 = selected, 0 = not). Subject to constraints: exactly 11 selected, formation position requirements met. This is fast but ignores pair interactions. (2) Pair-based MIP (advanced): Maximize Σ V(pair_ij) × X_ij, where X_ij = 1 if both players i and j are selected. Add overlap weighting: when a pair shares members with other selected pairs, their combined value increases. Subject to same formation constraints. This requires reformulating with both pair selection variables (X) and individual selection variables (Z) to maintain constraint satisfaction. (3) Validate by backtesting: form optimal teams for historical matches, compare to actual manager selections, measure prediction accuracy.

Results from Premier League backtesting: pair-based MIP produces lineups closer to actual manager selections than individual-based MIP. World Cup data (less rotation, fewer injuries) shows even stronger results. The pair-based method also predicts in-game metrics: 14% improvement on shot prediction, 13% on goals scored, 46% on passes.

Progression Levels

Diagnostic Tree

Coaching Cues

  • "4.5 million combinations, solved in 1.39 seconds. The math isn't the bottleneck — the context is."
  • "The best 11 against Man City isn't the best 11 against Burnley."
  • "This isn't an AI manager. It's a synergy calculator that shows you what you're losing when a player is out."

Common Errors

  1. Selection bias in pair scores: Players who start together frequently have inflated pair scores simply due to co-occurrence volume. Normalize by shared minutes.
  2. Ignoring formation constraints: An unconstrained MIP may select 5 midfielders and 4 forwards. Formation constraints are essential.
  3. Treating the MIP output as a lineup decision: The MIP identifies synergy-maximizing combinations. The manager adds context (form, fitness, opponent, morale) that the model doesn't capture.

Edges

⚡ Conventional Wisdom Is Wrong

Lineup Optimization Is a Constraint Satisfaction Problem, Not a Best-11-Players Problem

tactical-analysislineup-optimization-mip →

The best 11 individuals do not form the best team. Lineup optimization is a mixed-integer programming problem where player-role fit, pair synergy, positional coverage constraints, and game-model compliance interact. A mathematically optimal lineup may exclude the team's highest-rated individual player because their inclusion creates a worse collective configuration.

What most people do
Select the best player at each position independently, then hope the combination works.
What the best do
Model lineup selection as a constraint-satisfaction optimization: define the game model's positional requirements, include pair synergy scores as interaction terms, add hard constraints (minimum defensive coverage, minimum distribution quality from the back), and solve for the lineup that maximizes expected team output — not individual ratings.
Why it's an edge: A team with a 95th-percentile player who doesn't fit the system will underperform a team of 75th-percentile players who fit together. The interaction effects dominate the individual effects for most positions.
How to exploit: Build a lineup optimizer that takes individual ratings, pair synergy scores, and game model constraints as inputs. Run it before every match. Compare its output to the coach's selection — the divergences are the analytical insight.
James, University of Southampton, StatsBomb Innovation in Football Conference, 2019-10-30. Pair synergy scoring shows interaction effects.

Sources

  • James, University of Southampton, StatsBomb Innovation in Football Conference, YouTube, 2019-10-30 — presented MIP formulation for team selection maximizing pair synergy; 4.5M combinations solved in 1.39s; backtested against Premier League and World Cup lineups; 62% match outcome prediction accuracy