A Machine Learning Application for Football Players’ Selection

A Machine Learning Application for Football Players Selection

Onwuachu Uzochukwu C Department of Computer Sciences, Imo state University,

Owerri, Nigeria.

P. Enyindah

Department of Computer Sciences, University of Port Harcourt, Choba, Nigeria.

Abstract: The process of football team player selection is ultimately where the success of a team is determined. This is because the successful collection of individual talented players forms an effective team. In general, the selection of players in a football team is a decision made by the club on the basis of the best available information. This paper proposes a model that groups the attributes needed for player selection into four major categories which include the players technique, the players speed, the players physical status and the players resistance using neural network technique to determine these major attributes for each player. The system was developed and implemented using Matlab 2008. The result has shown that Neural Network is a good tool for selecting players in a football team.

Keywords: Football Team, Decision Support System, Player Selection, Neural Network.

1. INTRODUCTION

   The success of any football match lies in player selection which a difficult decision is making task for football managers. Football managers may need to use a decision support system to aid their decision making process. These attributes may include the players individual skills and performance statistics, combination of players, physical fitness, psychological factors, and injuries among others [1]. Some coaches may also use importance weights to determine the impact of each attribute. Importance weights are useful to the football managers since they indicate how the impact of a particular attribute relates to the probability of a successful outcome.

   Football is a team sport that is popular in almost every country in the world. The player selection process for professional football teams is crucial in the quest for winning. So much so that a wrong selection can cost football team the championship and even millions of dollars if the player turns out not living up to the teams expectations [22]. Traditionally, professional football teams use a variety of sports psychology assessments for evaluating players. There is no doubt that these assessments are of great benefit and are extremely useful when trying to form a winning football team. However, this is just one part of the big puzzle when trying to assess a players suitability for a team. The ability to select suitable players is indispensable when it comes to building an effective team [11].

   This paper makes use of Neural Network technique to build a decision support system for player selection in a football team. The system uses neural network to evaluate some attribute that makes up the four major categories to be considered by the football managers in selecting a player for a football team. A selected player will be judged based on these four categories which include the players technique, the players speed, the players physical status and the players resistance.

2. LITERATURE REVIEW

Katzenbach and Smith (1993) defined a team as a small number of people with complementary skills who are committed to a common purpose, performance goals, and approach for which they hold themselves mutually accountable [11]. A variety of approaches for the selection of team members have been proposed in the literature. Most of these studies have focused on the use of teams in business and industry. The business and industrys adoption of a teamwork methodology in the pursuit of cost effectiveness and greater innovation has spawned significant research [9][7][10].

Braha (2002) has proposed a team-building approach based on task partitioning by specifying task dependencies and partitioning the tasks among a number of teams [13].

Chen and Lin (2004) proposed a team member model for the formation of a multi-functional team in concurrent engineering. They used the analytic hierarchy process and MyersBriggs type indicators to model team member characteristics [14].

Gronau et al (2006) developed an algorithm to propose a team composition for a specific task by analyzing the knowledge and skills of the employees in the project management field [15]. Durmusoglu and Kulak (2008) proposed a team building process using axiomatic design principles. They proposed to establish teams by identifying the needed skills and preparing a skill development procedure to ensure maximum utilization of team members talents[20].

Feng, et al (2010) proposed a member selection method in cross functional teams where both the individual performance of the candidates and the collaborative performance between candidates were considered. Fuzzy set theory has also been used in the team member selection and team formation research [19].

Liang and Wang (1992) proposed integrating fuzzy logic into weighted complete bipartite graphs and developing a polynomial time algorithm for solving personnel placement[18].

DeKorvin, et al(2002) developed a model for the selection of personnel in multiple phase projects, which took into account the match between the skills possessed by each individual, the skills needed for each phase, and flexible budget considerations. They used the fuzzy construct of

Attack Awareness

Players

Resistance

Neural network

Defence Form

Tenacity Team work Acceleration

Figure1. Neural Network Model Architecture for Players Resistance Status

compatibility to measure the fit of a persons skill set to the goal set for each project phase in fuzzy environment [16] Boon and Sierksma (2003) formulated a linear optimization model to headhunt or scout a new team in soccer and

Response

Explosive power Dribble Speed Top Speed

Neural

Players Speed

network

volleyball by combining the qualities of the candidates and players with the functional requirements.[4].

Merigo´ and Gil-Lafuente (2011) analyzed the use of the ordered weighted averaging (OWA) operator in the selection of human resources in sport management. They used the Hamming distance, the adequacy coefficient and the index of maximum and minimum level to parameterize these decision-making techniques and select of a football player for a team.[12]

Ahmed et al (2011) considered the overall batting and bowling strength of a cricket team and proposed a constrained multi-objective optimization model for selection of the players on the team [2].

Askin and Sodhi (1994) have presented a novel method for organizing teams in concurrent engineering. They developedfive different criteria for team formation and discussed team training, leadership, and computer support issues [2].

Zakarian and Kusiak (1999) proposed an analytical model for the selection of multi-functional teams. They used the analytic hierarchyprocess and the quality function deployment method to prioritize team membership based on customer requirements andproduct specifications under fuzzy environment[21].

1. MATERIALS AND METHOD

   Neural network technique was used to develop a model for player selection, in this model different attributes needed for player selection is analyzed into four major categories using a neural network model and this major categories which include the players technique, the players speed, the players physical status and the players resistance. The data used in this paper was collected from www.pesstatsdatabase.com/PSD/playerClassic.php?id=157 98&club=0 These major categories will be used to make the final decision on which player to be selected for a football team.Figure1, figure 2, figure 3. and fig 4 shows

   Figure 2. Neural Network Model Architecture for Players Speed Status

   Players Physical

   Neural network

   Body Balance Stamina Jumping Kicking Power Injury Tolerance

   Figure 3. Neural Network Model Architecture for Players Physical status

   Neural network

   Attack Defence

   Players Technique

   Header Accuracy Dribble Accuracy Short Pass Accuracy Short Pass Speed Long Pass Accuracy Long Pass Speed Shot Accuracy Place Kicking Swerve

   Ball Control

   Goal Keeping Skills Weak Foot Accuracy Weak Foot Frequency

   Figure 4. Neural Network Model Architecture for Players Technique

   The feed forward algorithm was used to calculate the optimal weights of the individual attributes that make up these categories. The mathematical models for the feed forward algorithm are as follows:

   Input j x j yi wij 1

   yiis the generated output and wi j represents weights

   1

   the neural network model architecture for calculating the players technique, the players speed, the players physical status and the players resistance respectively.

   f (x)

   1 exj

   …………………………………..2

   f (x) is a sigmoid that is used as the activation function

   Error Tk Ok 3

   Tk is the observed (True) output while Ok is the calculated (actual) output

   The error in the output layer is calculated by using the formula in equation 3.4

   k ok (1 – ok )(Tk ok ) 4

   Where Ok is the calculated (actual) output expressed in equation 3.5

   1. EXPERIMENTS AND RESULTS

      The simulation was done using Matlab 2008. Figure 5 shows the Neural Network fitting tool for data selection. This Interface helps in collection of input data and the target data from the work space.

      Ok

      1

      1 e xk

      ……………………………………….5

      Tk is the observed (True) output

      The back propagation error in the hidden layer is calculated by using the formula in equation 3.6

      j oj (1 – oj )k * wjk 6

      k

      Where wjk is the weight of the connection from unit j to unit k in the next layer and k is the error of unit k.

      The weight adjustment formula in equation (3.7) is used to adjust the weights to produce new weights which are fed back into the input layer.

      Wnew Wold * * input 7

      Where is a constant called the learning rate. The learning rate takes value between 0 and 1.

      A. THE CONDITION FOR PLAYER SELECTION

      The Average Status = (Physical status + Technique status + Speed status + Resistance status) / 4

      X < 50

      Average stat (X) = X = 50

      X > 50

      From the first condition, if the players average status is below 50 then the player rating is below average and with this information the coach will reject the player. If the players average status is equal to 50 then the player rating is average and with this information the coach can choose the player but he must not be in the starting eleven, which means he/she can be in the bench. If the players average status is above 50 thenthe player rating is above average and with this information the coach can choose the player in the starting eleven.

      Figure 5. Neural Network Fitting Tool for Data Selection

      Figure 6 shows the Neural Network fitting tool for selection of network size. This interface gives the user the opportunity to select the number of neuron in the networks hidden layer. The user can return to this panel and change the number of the neuron if the network does not perform well after training.

      Figure 6. Neural Network Fitting Tool for Network Size Selection

      Figure 7.shows the Neural Network training. The Neural Network model was trained using Levenberg-Marquardt back propagation. The network is trained to fit the inputs and the target. This means that neural network map between a data of numeric inputs and a set of numeric targets. Training automatically stops when generalization stops improving as indicated by the increase in the mean square error of the validation samples.

      Figure 7. Neural Network Training Tool

      The neural network fitting tool will help in training network and evaluation its performance using mean square error and regression analysis. Training multiple times will generate different results due to different initial condition and sampling. Figure 8.shows the result of the trained Network.

      Fig 8. Neural Network Fitting Tool for Displaying the Result of the Trained Network.

      Table 1. shows the comprehensive summary of the results from the developed system. The result shows the different player capabilities, and the team managers select their players base on the generated result. The table also displays the average result based on the four major categories for easy clarification and selection.

      Table 1: The Comprehensive Summary of the Results from the developed system

      Players Name	Technique status	Speed status	Physical status	Resistance status	Average status	Player rating	Decision
      Lionel Messi	85.1	92.1	62.9	28.2	67.075	Above Average	Select
      Christiano Ronaldo	82.3	92.4	78.3	33.4	71.6	Above Average	Select
      Arjen Robben	82.2	93.1	64.6	32.8	68.175	Above Average	Select
      Andres Iniesta	88.4	76.3	62.8	59.3	71.7	Above Average	Select
      Luis Suarez	79.6	83.7	79.4	43.3	71.5	Above Average	Select
      Franck Ribery	86.2	88.8	62.3	30.5	66.95	Above Average	Select
      Eden Hazard	85.1	88.1	63.5	32.6	67.325	Above Average	Select

      Table 2: The Comprehensive Summary of the online Results on 16th Dec 2014

      Players Name	Technique status	Speed status	Physical status	Resistance status
      Lionel Messi	86	93	62	27
      Christiano Ronaldo	81	93	79	32
      Arjen Robben	83	93	64	32
      Andres Iniesta	89	75	63	59
      Luis Suarez	79	83	79	42
      Franck Ribery	85	89	62	29
      Eden Hazard	84	89	64	32

      100

      80

      60

      40

      20

      0

      Franck Ribery online

      value

      Franck Ribery system value

      Technique Speed Physical Resistance

      status status status status

      Figure 10: A Line Graph That Shows the Relationship Between Online and System Result of Frank Ribery

      100

      80

      60

      20

      0

      Christiano Ronaldo

      online value

      Christiano Ronaldo system value

      Technique Speed Physical Resistance

      status status status status

      Figure 11: A Line Graph That Shows the Relationship Between Online and System Result of Christiano Ronaldo

      100

      80

      60

      40

      20

      0

      Andres Iniesta online

      value

      Andres Iniesta system value

      status

      status

      Physical Resistance

      Speed

      status

      Technique

      status

      Figure 12: A Line Graph That Shows the Relationship Between Online and System Result of Andres Iniesta

      100

      80

      60

      40

      20

      0

      Arjen Robben online

      value

      Arjen Robben system value

      Technique Speed Physical Resistance

      status status status status

      Figure 13: A Line Graph That Shows the Relationship between Online and System Result of Arjen Robben

      100

      80

      60

      40

      20

      0

      Lionel Messi online

      value

      Lionel Messi system value

      Technique Speed Physical Resistance

      status status status status

      Figure 14: A Line Graph That Shows the Relationship Between Online and System Result of Lionel Messi

      100

      80

      60

      40

      20

      0

      Luis Suarez online

      value

      Luis Suarez system value

      Technique Speed Physical Resistance

      status status status status

      Figure 15: A Line Graph That Shows the Relationship Between Online and System Result of Luis Suarez

      100

      80

      60

      40

      20

      0

      Eden Hazard online

      value

      Eden Hazard system value

      Technique Speed Physical Resistance

      status status status status

      Figure 16: A Line Graph That Shows the Relationship Between Online and System Result of Eden Hazard

   2. RESULT DISCUSSION

      Table 1.shows the comprehensive summary of the results from the developed system. The result shows the different player capabilities. Based on this generated result, players are being selected by the team manager. The table also displays the average result based on the four major categories for easy clarification and selection. Table 2.displays the comprehensive summary of the online results on 16th Dec 2014. The reason for this result is to compare the accuracy of the developed system with the existing system. Figures 10, 11, 12, 13, 14, 15 and 16 show a line graph displaying the relationship between online and system result ofFrank Ribery,Christiano Ronaldo, Andres Iniesta, ArjenRobben, Lionel Messi, Luis Suarez and Eden Hazard respectively. From the graphs, it can be seen that neural network can be used to predict player performance in a team with minimum error. These results have shown

      that Neural Network is a good tool in building a decision support system for a football team manager.

   3. CONCLUSION

Neural Network has been proven in the paper to be a good tool for building a decision support system for a football team manager. There are some attributes that a football player may have which cannot be neglected when it comes to choosing a rightful player for a football team. This system has employed the idea of neural network in considering this large amount of attributes needed in secting the rightful player for a football team. The result generated from the system has shown that neural network technique can help the football managers in player selection for a football team.

REFERENCES

1. Arnason, A., Sigurdsson, S. B., Gudmundsson, A., Holme, I., Engebretsen, L., & Bahr, R. (2004).Physical fitness, injuries, and team performance in soccer.Medicine &Science in Sports & Exercise, 36, 278285.

2. Ahmed, F., Deb, K., & Jindal, A. (2011).Evolutionary multi- objective optimization and decision making approaches to cricket team selection.In Proceedings of the Second International Conference on Swarm, Evolutionary, and Memetic Computing. Berlin, Heidelberg: Springer- Verlaghttp://dx.doi.org/10.1007/978-3-642- 27242-4_9.

3. Askin, R. G., &Sodhi, M. (1994).Organization of teams in concurrent engineering.In R. D. Dorf& A. Kusiak (Eds.), Handbook of design, manufacturing, and automation (pp. 85105). New York: John Wiley & Sons

4. Boon, B. H., &Sierksma, G. (2003). Team formation: Matching quality supply and quality demand. European Journal of Operational Research, 148, 277292.

5. Katzenbach, J. R., & Smith, D. K. (1993).The wisdom of teams. Boston: Harvard Business School Press.

6. Kinlaw, D. C. (1991).Developing superior work teams. San Diego: Lexington Books

7. Chen, Y. L., Cheng, L. C., & Chuang, C. N. (2008).A group recommendation system with consideration of interactions among group members.Expert Systems with Applications, 34, 20822090.

8. English, A., Griffith, R. L., &Steelman, L. A. (2004). Team performance: The effect of team conscientiousness and task type. Small Group Research, 35(6), 643665

9. Kirkman, B. L., Rosen, B., Tesluk, P. E., & Gibson, C. B. (2004). The impact of team empowerment on team performance: The moderating role of face-to-face interaction. Academy of Management Journal, 47(2), 175192.

10. Mannix, E., & Neale, M. (2005). What differences make a difference? The promise and reality of diverse teams in organizations. American Psychological Society, 6(2), 3155

11. Boon, B. H., &Sierksma, G. (2003). Team formation: Matching quality supply and quality demand. European Journal of Operational Research, 148, 277292.

12. Merigo´ , J. M., & Gil-Lafuente, A. M. (2011). Decision- making in sport management based on the OWA operator. Expert Systems with Applications, 38, 1040810413.

13. Braha, D. (2002). Partitioning tasks to product development teams. Paper presented at the International Design Engineering Technical Conferences of American Society of Mechanical Engineers (DETC02 ASME).

14. Chen, S. J., & Lin, L. (2004).Modeling team member characteristics for the formation of a multifunctional team in concurrent engineering.IEEE Transactions on Engineering Management, 51(2), 111124.

15. Gronau, N., Fro¨ming, J., Schmid, S., &Ru¨ ssbu¨ ldt, U. (2006). Approach for requirement oriented team building in industrial processes. Computers in Industry, 58(2), 179187.

16. DeKorvin, A., Shipley, M., &Kleyle, R. (2002). Utilizing fuzzy compatibility of skill sets for team selection in multi- phase projects. Journal of Engineering and Management, 19, 307319

17. TechnoDereli, T., Baykasoglu, A., & Das, S. (2007). Fuzzy quality-team formation for value added auditing: A case study. Journal of Engineering and Technology Management, 24, 366394.

18. Yaakob, S. B., &Kawata, S. (1999). Workers placement in an industrial environment. Fuzzy Sets and Systems, 106, 289297.

19. Liang, G. S., & Wang, M. J. J. (1992).Personnel placement in a fuzzy environment. Computers and Operations Research, 19, 107121.

20. Feng, B., Jiang, Z. Z., Fan, Z. P., & Fu, N. (2010).A method for member selection of cross-functional teams using the individual and collaborative performances.European Journal of Operational Research, 203, 652661.

21. Durmusoglu, M., & Kulak, O. (2008). A methodology for the design of office cells using axiomatic design principles.

    Omega, 36, 633652

22. Zakarian, A., &Kusiak, A. (1999).Forming teams: An analytical approach. IIE Transactions, 31, 8597.

23. Madjid Tavana, Farshad Azizi, Farzad Azizi, Majid Behzadian. (2012), A fuzzy inference system with application to player selection and team formation in multi- player sports, Sport Management Review, jour n al homepag e: www .elsevier .c om /lo cate/s m r