Shot Prediction in the Attacking Third Based on Spatio-temporal Features: A Dynamic Time-series Model Approach

The attacking third plays an important role in football, as it is the area where goal-scoring chances typically emerge and often determines the outcome of a match. This study aims to construct an optimal model for accurately predicting shot probabilities in the attacking third, focusing on the impact of reception events on shooting behavior. This study includes 52 matches from the FIFA U17 World Cup Indonesia 2023. The study constructs a predictive model for shot opportunities following ball receptions in the attacking third, employing five dynamic time-series algorithms: LSTM, BiLSTM, GRU, 1D CNN, and CNN+LSTM, based on match event location data. The performance of these models was evaluated using confusion matrices and AUC metrics, with LSTM achieving the highest recall rate of 87.1%, effectively identifying ball reception and shot events in the attacking third. The 1D CNN model demonstrated the highest AUC value of 0.933, highlighting its superior ability to differentiate between “shot formation” and “non-shot formation” events. The comprehensive approach enables a more precise evaluation of potential shot opportunities, providing valuable tactical insights for teams. By comparing different algorithms, this research contributes to the development of more efficient and accurate models for football data analysis, advancing modern football performance analysis methodologies.