A Markov Decision Process (MDP) specifically for modeling buildup sequences — the structured phase where a defender has the ball, no offensive pressure is applied, and the team attempts to progress into the final third. Unlike general xT or EPV models that cover all possessions, this model is conditioned on the defensive setup being faced (high block, low block, left-forcing, right-forcing). Using StatsBomb 360 data, the model incorporates both ball location (30 zones) and reachable teammate information (4 binary features per state: is there a reachable teammate to the left, right, in front, or behind the ball carrier). This produces a separate behavioral model per defensive setup per team.
(1) Extract buildup sequences: filter possessions where a defender has the ball with no offensive pressure. (2) Classify defensive setup from 360 data: high block vs. low block (based on defensive line height); left-forcing vs. right-forcing (based on defensive shape asymmetry). (3) Build one MDP per setup: states = 30 pitch zones; actions = move-to (with intended destination) or shoot; transition probabilities extracted from event data; reward = 1 for reaching the final third. (4) Extended MDP: add reachable teammate info as 4 binary features per state — teammate is "reachable" if no opponent is within 1.5m of them AND no opponent is close enough to the passing lane to intercept. (5) Extract most likely sequences per team per defensive setup to identify preferred buildup patterns.
Key data requirement: StatsBomb 360 data (player position snapshots for every action) is essential for defensive setup classification and reachable teammate modeling. Standard event data alone is insufficient.