![](bg8.png)
layered network suggests that GoogLeNet is indeed
able to catch some higher-order patterns, in which the
3-layered net does not succeed.
Then remains the question how bad an accuracy
of 67.1% actually is. As a starting point, it is signifi-
cantly higher than the 57.3% accuracy of the k-nearest
neighbors baseline. But what would the performance
of the best performing human or artificial classifier
be for the current problem? Would an accuracy of,
say, 90% be realistic? This would probably not be the
case. Samples were extracted already 10 seconds be-
fore the occurrence of goal-scoring opportunities. A
lot can happen in the 10 second interval to the actual
shot towards the goal.
5 CONCLUSION
To conclude, the results suggest that convolutional
neural networks are capable of predicting goal-
scoring opportunities to a certain extent. There are
a couple of ways how the performance of the convo-
lutional neural networks could be improved. Using a
larger dataset would probably stimulate the convolu-
tional neural networks to learn higher-order patterns
instead of more superficial ones. A higher number of
classes can be used to detect more events in soccer,
or input images can be weighted differently by taking
their temporal location to events into account. As an
alternative to using point images, a graph representa-
tion of the position data could be used as input to neu-
ral networks. Finally, using a more specialized con-
volutional neural network, possibly combined with
a recurrent neural network architecture, could yield
higher accuracies, as would using an ensemble of sev-
eral classifiers.
REFERENCES
Barris, S. and Button, C. (2008). A review of vision-
based motion analysis in sport. Sports Medicine,
38(12):1025–1043.
Bengio, Y., Simard, P., and Frasconi, P. (1994). Learning
long-term dependencies with gradient descent is diffi-
cult. IEEE transactions on neural networks, 5(2):157–
166.
Brooks, J., Kerr, M., and Guttag, J. (2016). Using machine
learning to draw inferences from pass location data
in soccer. Statistical Analysis and Data Mining: The
ASA Data Science Journal, 9(5):338–349.
Cover, T. and Hart, P. (1967). Nearest neighbor pattern clas-
sification. IEEE transactions on information theory,
13(1):21–27.
Deng, J., Dong, W., Socher, R., Li, L.-J., Li, K., and Fei-Fei,
L. (2009). Imagenet: A large-scale hierarchical image
database. In Computer Vision and Pattern Recogni-
tion, 2009. CVPR 2009. IEEE Conference on, pages
248–255.
Grunz, A., Memmert, D., and Perl, J. (2012). Tactical pat-
tern recognition in soccer games by means of spe-
cial self-organizing maps. Human movement science,
31(2):334–343.
Jacobs, R. A. (1988). Increased rates of convergence
through learning rate adaptation. Neural networks,
1(4):295–307.
Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J.,
Girshick, R., Guadarrama, S., and Darrell, T. (2014).
Caffe: Convolutional architecture for fast feature em-
bedding. In Proceedings of the 22nd ACM interna-
tional conference on Multimedia, pages 675–678.
Knauf, K., Memmert, D., and Brefeld, U. (2016). Spatio-
temporal convolution kernels. Machine Learning,
102(2):247–273.
Kohonen, T. and Somervuo, P. (1998). Self-organizing
maps of symbol strings. Neurocomputing, 21(1):19–
30.
Krizhevsky, A., Sutskever, I., and Hinton, G. E. (2012). Im-
agenet classification with deep convolutional neural
networks. In Advances in neural information process-
ing systems, pages 1097–1105.
LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard,
R. E., Hubbard, W., and Jackel, L. D. (1990). Hand-
written digit recognition with a back-propagation net-
work. In Advances in neural information processing
systems.
Memmert, D., Lemmink, K. A., and Sampaio, J. (2016).
Current approaches to tactical performance analyses
in soccer using position data. Sports Medicine, pages
1–10.
Moody, J., Hanson, S., Krogh, A., and Hertz, J. A. (1995).
A simple weight decay can improve generalization.
Advances in neural information processing systems,
4:950–957.
Nair, V. and Hinton, G. E. (2010). Rectified linear units
improve restricted boltzmann machines. In Proceed-
ings of the 27th International Conference on Machine
Learning (ICML-10), pages 807–814.
Nesterov, Y. (1983). A method of solving a convex pro-
gramming problem with convergence rate o (1/k2). In
Soviet Mathematics Doklady, volume 27, pages 372–
376.
Sutskever, I., Martens, J., Dahl, G. E., and Hinton, G. E.
(2013). On the importance of initialization and mo-
mentum in deep learning. ICML (3), 28:1139–1147.
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S.,
Anguelov, D., Erhan, D., Vanhoucke, V., and Rabi-
novich, A. (2015). Going deeper with convolutions.
In Proceedings of the IEEE Conference on Computer
Vision and Pattern Recognition, pages 1–9.
Zeiler, M. D. and Fergus, R. (2014). Visualizing and under-
standing convolutional networks. In European Con-
ference on Computer Vision, pages 818–833. Springer.
Using Deep Convolutional Neural Networks to Predict Goal-scoring Opportunities in Soccer
455