Table 4. Comparison with existing literature (previously published models were trained on shot event only).

Research	Model	ROC-AUC	Evaluation Criteria
Eggels (2016) [10]	Random Forest	0.775	Average cross-validation performance with event-log+player-tracking data from 5020 professional games
	Decision Trees	0.777
	Logistic Regression	0.785
	ADA Boost	0.692
Anzer and Bauer (2021) [21]	Random Forest	0.794	Unseen 20% a test data split from an event-log data set consist of 105,207 shots in the German Bundesliga
	GBM	0.822
	Logistic Regression	0.807
	ADA Boost	0.817
Haaren (2021) [22]	Boosting Machine	0.793	Unseen event-log test data consist of 38,737 shots in European football leagues
Mead (2023) [11]	XGBoost	0.800	30% test data split from WYSCOUT data set [44, 45]
The present work			Average performance from 100 rounds of five-fold cross-validation + average performance with unseen unbalanced and balanced testing data from 500 models developed
Validation Data	Random Forest (3 events) 1	0.833
Unbalanced testing	Random Forest (4 events) 2	0.827
Balanced testing	Random Forest (4 events) 3	0.826

¹ The highest mean ROC-AUC value with validation data was obtained when shot event information and two preceding events information (3 events) were included (0 ≤ n ≤ 2 events)

² The highest mean ROC-AUC value with unseen unbalanced testing data was obtained when shot event information and three preceding events information (4 events) were included (0 ≤ n ≤ 3 events)

³ The highest mean ROC-AUC value with unseen balanced testing data was obtained when shot event information and three preceding events information (4 events) were included (0 ≤ n ≤ 3 events)