Fig 1.
One of the mechanisms of SPP identifies and removes patterns that do not contribute to the model before performing the optimization. For example, if pattern t does not satisfy the pruning criterion specified in [34], the sub-tree below pattern t is deleted.
Table 1.
Prior studies that have applied sequential pattern mining techniques in sport.
Table 2.
Unique events in the original XML data.
Events prefixed by “O-” are performed by the opposition team; those that are not are performed by the team.
Fig 2.
The photographs used as the original images are listed in parentheses. All of them are licensed under the unsplash.com license (https://unsplash.com/license). Top left: Kick at goal (https://unsplash.com/photos/xJSPP3H8XTQ); Bottom left: Line-out (https://unsplash.com/photos/CTEvFbFpVC8); Center top: Kick restart/Kick-off (https://unsplash.com/photos/OMdge7F2FyA); Center bottom: Scrum (https://unsplash.com/photos/y5H3_7OobJw); Top Right: Line break (https://unsplash.com/photos/XAlKHW9ierw); Middle Right: Beginning of a phase (https://unsplash.com/photos/fqrzserMsX4); Bottom Right: Breakdown (https://unsplash.com/photos/WByu11skzSc).
Fig 3.
Converting XML files into labeled sequences.
Illustration of the procedures and specified rules to delimit the raw XML data files into passage of play event sequences labeled with scoring or conceding outcomes.
Table 3.
Event lists for the original, scoring and conceding datasets.
Fig 4.
Illustration of dataset creation and experimental approach.
Illustration of the procedures to create the datasets from the original delimited dataset to be used in the experiments and to compare the unsupervised and supervised SPM methods.
Fig 5.
Sequence length distributions.
Distribution of sequence lengths by points-scoring outcome for the scoring and conceding datasets. Sequence length is defined as the number of events in each sequence (excluding the outcome label).
Table 4.
Descriptive statistics for the scoring and conceding datasets.
Table 5.
Top five SPP-obtained patterns that discriminated the most between scoring and non-scoring outcomes.
Table 6.
Top five SPP-obtained patterns that discriminated the most between conceding and non-conceding outcomes.
Table 7.
Top five PrefixSpan-obtained patterns with the largest support: Scoring+1 dataset.
Table 8.
Top five PrefixSpan-obtained patterns with the largest support: Conceding+1 dataset.