SMART FOOTBALL : DEVELOPMENT OF SIMULATION MODEL FOR ATTACKING SCHEME ASSESSMENT BASED ON COORDINATE AND PLAYER’S SKILL WITH LOGISTIC REGRESSION

• Main Author: Immanuel Harnadi, Steve
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/81835
• Institution: Institut Teknologi Bandung

Local players’ quality in Indonesia is still behind the naturalized players brought in to strengthen the Indonesian National Team, so that training’s quality in the clubs of Liga 1 Indonesia should be improved. Various researches to support the goal has been already achieved, however there are still basic problems such as for example (1) important data should be inputted manually, (2) unstandardized data standard across different researches, (3) game scheme evaluation only based on passing action, (4) game scheme evaluation depends on limited data, and (5) player’s skill ability not yet included in game scheme evaluation. Refer to that, this research aims for building simulation model for attacking scheme assessment only based on player’s coordinate (SPADL) and player’s skill score (sofifa). This simulation model is built according to 2 types of simulation model, i.e. difficulty value simulation model and player skill probability simulation model, which are applied on 3 player’s action (passing action (xpass model), shooting action (xgoal model), dribbling action (xdribble model)). Both Difficulty Value Simulation Model and Player Skill Probability simulation model are built on Logistic Regression algorithm. Multiplication result from both simulation models is considered as success probability number for a player doing certain actions in game scheme, which adopts Stochastic Process concept. This simulation model consisted of 3 player’s action is then evaluated in various testing scenario such as simulated game scheme scenario, real game scheme scenario in real world, and model usage for detecting potential players compared to VAEP method as state-of-the-art. Evaluation result shows that xpass model and xgoal model relatively yield good performance in most of evaluation’s scenario, except for xdribble model still needs a improvement. For detecting potential players overally, this simulation model yields far better result than VAEP method if 10 highest ranking players are sorted excluded another factors such as goal’s ratio, assist ratio, or goal and assist’s ratio within 90 minutes match in 2 big competitions.