Fig 1.
Performance comparison of the existing machine learning-based models for soccer match outcome prediction.
Table 1.
Accuracy of predicting football match outcome using deep-learning based models.
Table 2.
Summary statistics about the dataset.
Table 3.
High-level summary of the information (feature) available in the dataset.
Fig 2.
Proposed architecture for the SoccerNet for predicting match winners.
FCL: Fully connected layer. GRU: Gated Recurrent Unit.
Fig 3.
The training and validation losses from the outer folds of the experiment.
Table 4.
Number of parameters in each layer of our proposed model.
Table 5.
Performance of ML models on test set for both tabular and sequential (position-agnostic and position-aware) approaches on test set.
Table 6.
The effect of different player positions on the outcome.
Table 7.
The effect of different match segments on the outcome considering test set using the final model (based on GRU Net).
Fig 4.
Performance of the proposed model considering players’ position and match segment.