Davis Lab Tree Ensembles Validate Throw-in Setups Within 10 Actions of Goal

KU Leuven researchers trained tree ensemble models on 2022 World Cup event data to quantify the value of intentional throw-ins. The models simulate sequences after the ball exits play near the opponent goal and output expected goal increments. Clubs including Anderlecht integrated these outputs into pre-match planning for the 2024-25 season.

Real-time optical tracking systems now feed the same variables Davis used—player velocities, pitch zones, and action counts—directly into broadcast overlays. This replaces static xG with live recovery probabilities displayed after each throw-in. Three Bundesliga sides adopted the tactic at rates above 12 percent in opening matches.

Standardization of tracking data formats remains the next bottleneck. Davis's open tools allow clubs to align Opta and Second Spectrum feeds without proprietary mappings, cutting integration time from weeks to days. Broadcast partners plan to surface these probabilities in 2025 World Cup qualifiers.

Operational impact appears in set-piece staff reallocations and scouting filters that now weight throw-in recovery metrics alongside traditional passing accuracy.