See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/276485002

English Premier League (EPL) Soccer Matches Prediction using An Adaptive

Neuro-Fuzzy Inference System (ANFIS)

Article · April 2015

DOI: 10.14738/tmlai.32.1027

CITATIONS READS

2 2,516

2 authors:

Frank Amadin Obi J.C

University of Benin 8 PUBLICATIONS   43 CITATIONS   
29 PUBLICATIONS   18 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Multi-tenancy View project

Automation of Students Results Processing View project

All content following this page was uploaded by Frank Amadin on 22 June 2016.

The user has requested enhancement of the downloaded file.

 English Premier League (EPL) Soccer Matches Prediction using
An Adaptive Neuro-Fuzzy Inference System (ANFIS) for
Amadin, F. I1 and Obi, J.C.2
Department of Computer Science, University of Benin, Benin City. Nigeria.
[email protected]; [email protected]

ABSTRACT

Prediction of English Premiership League (EPL) matches has been on the heart and minds of researcher
over the pass decades, but none has sufficiently introduced and Adaptive Neuro-Fuzzy Inference System
(ANFIS) approach for these prediction which has served as the focal aim of this research paper using
seven premier league teams and nine linguistic values. Matric Laboratory (MATLAB) 7.0 served as the
tool of implementation highlighting various views. The ANFIS training was successful completed at
epoch 2 and having an error of 1.41237e-006. The model was further used to predict the outcome of 7
matches with a successful rate of 71%.
Keywords: ANFIS, Premiership, Soccer, Predication, Layers

1 Introduction
It is no doubt that the most viewed sport in the world is soccer having above 10 billon fan all over the
world (Tony, 2014). The English league is the most watched soccer league all over the world having
more than 2 billion fans (Jonathan, 2014 and Tony, 2014). It comprises of 20 teams and each team plays
38 match in a season which spans ten months. A team plays 19 matches at home (at the city in which
the club originates) and 19 match away (Jonathan, 2014). A win for any match regardless of home or
away is 3 points a draw is 1 point while a loss earn no point. At the end of each season the 3 teams at
the lower end (bottom) of the league are relegated to a lower division (Tony, 2014). The outcome of any
match is a win, a draw or a loss (Jonathan, 2014 and Tony, 2014). To calculate the number of games
played in a season using the hand shaking lemma we represent each team as a node and a matched
played as an edge. There are two parallel edges between each node (one for home match and the other
for away match). So, the number of game played = (38*20) / 2 =380 matches. Since the edge contribute
twice to the degree of the graph.

2 Review of Related Literature

In predicting the outcome of a soccer match, various approaches have been used some of which include
Statistical method, Probabilistic method, Bayesian network, Multilayer perceptrons.
Yue (2003) used multilayer perceptron model to predict the outcome of a soccer match. The neural
network had 13 inputs, 3 hidden layers and 1 output layer. The network was feed with the most recent 5
matches of team A and team B. The model was used to predict the outcome of 9 matches and 4 out of
the nine matches gave accurate prediction. Based on the efficiency of the result, the model was re-

DOI: 10.14738/tmlai.32.1027
Publication Date: 28th April, 2015
URL: http://dx.doi.org/10.14738/tmlai.32.1027
 Transactions on Machine Learning and Artificial Intelligence Volume 3 , Issue 2, April 2015

implemented with two additional hidden layers (a layer for the home team with a weight of 1 and the
other for the position of the team on the table of ranking) which was utilized in predicting nine matches,
with an 80% accuracy Aditya et al (2013) used a Multi Nominal Logic Regression (MNLR) and Support
Vector Machines (SVM) to predict the outcome of a match utilizing 3 fundamental approaches. In their
first approach Multi nominal Logistic was enacted for the training phase in which performance
optimization metrics derived from current matches were considered, rather than taking the average of
previous matches and during testing they predicted the outcome between two teams using the number
of matches past played. In the second approach, they trained in the same way they later test the data
and instead of using the feature vector as the performance metric vector corresponding to the current
match, they used Key Performance Parameters (KPP). The third approach tried to find a global set of
parameter which was independent of the competing teams.
Ian and Phil (2013), forecasted international; soccer matches result using bivariate discrete distribution.
They collected a total of 8,735 international soccer results from two main sources. The data for the
period 1993-2001 was obtained from the archive of International Soccer Results (ISR) and the data for
the period 2001 to 2004 were obtained from the Record Sport Soccer Statistics Foundation (RSSSF)
archive. Data on the Federation of International Football Association (FIFA) world rankings were
collected from the FIFA website for each month during the years from 1993 to 2004. The model was
based on copula functions. The copula regression model forecasts 41.8% of the results correctly.
From the review of related literature An Adaptive Neuro-Fuzzy Inference System (ANFIS) approach has
not been adopted previously for EPL matches prediction which is serving as the focal point of our
research.

3 The ANFIS Model for Predicting the Outcome of English Premiership

League (EPL) Soccer Match
The Adaptive Neuro Fuzzy Inference System (ANFIS) is one of the many hybrids of neural and fuzzy
system. ANFIS combines the learning capability of the neural network and the explanatory power of the
fuzzy system. The ANFIS architecture uses the sugeno type inference system Jang (1991). An example of
the sugeno inference is given in equation 1:

If x is A and y is B, then f1 = p1x + q1y + r1 (1)

Where x and y is the input variable and A and B is the fuzzy set of linguistic variables and q, p and r, are consequent
parameters.
ANFIS ARCHITECTURE

Figure 1: ANFIS ARCHITECTURE

Copyright © Society for Sci ence and Education United Kingdom 35

Amadin, F. I and Obi, J.C.; English Premier League (EPL) Soccer Matches Prediction using An Adaptive Neuro-Fuzzy
Inference System (ANFIS) for. Transactions on Machine Learning and Artificial Intelligence, Volume 3 No 2 April
(2015); pp: 34-40
The ANFIS architecture in Figure 1 comprises of varied layers which includes:
a. Layer 1 (Input Layer): this is the first layer it is called the input layer. Each node
generates the membership grades of a linguistic label (Jang, 1993). An example of a
membership function is the generalized bell function is given in equation 2:
1
 Ai ( x ) 
( x  ci ) 2 bi (2)
1[ 2
]
ai
Where {a, b, c} is the parameter set and x stands for the individual value. As the values of the
parameters change, the shape of the bell-shaped function varies. Parameters in that layer are called
premise parameters.
b. Layer 2 (Fuzzification layer): The second layer of the Adaptive Neuro Fuzzy Inference
System (ANFIS) model is fuzzification layer it is also called the membership function
layer. In this layer the output of the first layer is mapped to a fuzzy set using bell-shaped
membership function (Jang, 1997 and Nauck et al., 1997). The Bell membership
function provides smooth and non-linear functions that can be used by the learning
systems. Each node calculates the firing strength of each rule using the min or prod
operator show in equation 3. In general, any other fuzzy AND operation can be used.

O2,i  wi   Ai ( x)  Bi ( x) i  1, 2 (3)

c. Layer 3 (Rule Layer): The third layer is the rule layer. Each neuron in this layer
corresponds to a single first-order Sugeno fuzzy rule receiving signal from the
membership function layer and computes the truth value of the rule (Jang, 1993 and
1997. The nodes calculate the ratios of the rule’s firing strength to the sum of all the
rules firing strength using the equation 4. The result is a normalised firing strength.
wi
O3,i  wi  i  1, 2 (4)
w1  w2
d. Layer 4 (Normalization layer): The fourth layer is the normalization layer. Each neuron
in this layer receives signals from all rule neurons in the third layer, and calculates the
normalized firing strength of a given rule. The nodes compute a parameter function on
the layer 3 output using equation 5. Parameters in this layer are called consequent
parameters.
O4,i  wi fi  wi ( pi x  qi y  ri ) (5)

e. Layer 5 (Defuzzification layer): The fifth layer is the defuzzification layer. Each neuron in
this layer is connected to the respective normalization neuron in the fourth layer (Jang,
1997 and Nagnevitsky, 2002). It is a single node that aggregates the overall output as
the summation of all incoming signals using equation 6

O5,1  overall output   wi f i (6)

i

URL:http://dx.doi.org/10.14738/tmlai.32.1027 36
 Transactions on Machine Learning and Artificial Intelligence Volume 3 , Issue 2, April 2015

3.1 Generating Dataset

In generating the dataset we relied on 5 factors which then constituted the ANFIS parameters which was
then used to predict the outcome of the match, they are:

 2 of the last most recent match played by team A

 2 of the last most recent match played by team B
 The point of both teams in the table of ranking
 Their popularity
 Home advantage.
The first 2 parameters (2 of the last most recent match played by team have 9 linguistic variable (WW,
WD, WL, DW, DD, DL, LW, LD, LL) while the point of both teams and the popularity of the teams have 3
parameters (highA, same, lowA) and the home advantage have 2 linguistic variable (homeA and awayA).

3.2 Variable Used In the Model

Table 1, shows, six linguistic variables and six linguistic values associated with the Anfis proposed
architecture for predicting the EPL results and matches
Table 1: Fuzzy Linguistic Variables and Values

Position Club Last Two Match Point

Played
1 Chelsea WD 33
2 Manchester City WW 25
3 Southampton WW 22
4 Manchester United WW 21
5 West ham United DW 20
6 Arsenal LW 20

4 Stimulations and Experiment

The stimulation was carried out using MATLAB Fuzzy logic toolbox 2007 and the model was used to
predict the outcome of 7 matches. The various simulation modules are specified from Figure 2, 3, 4 and
5 respectively

Figure 2: Training Error

Copyright © Society for Sci ence and Education United Kingdom 37

Amadin, F. I and Obi, J.C.; English Premier League (EPL) Soccer Matches Prediction using An Adaptive Neuro-Fuzzy
Inference System (ANFIS) for. Transactions on Machine Learning and Artificial Intelligence, Volume 3 No 2 April
(2015); pp: 34-40

Figure 3: Surface View Prediction 1

Figure 4: Surface View Prediction 2

Figure 5: Surface View Prediction

The ANFIS training was completed at epoch 2 and having an error of 1.41237e-006. The model was
further used to predict the outcome of 7 matches and 5 outcomes out of the matches were accurately
predicted.

URL:http://dx.doi.org/10.14738/tmlai.32.1027 38
 Transactions on Machine Learning and Artificial Intelligence Volume 3 , Issue 2, April 2015

4.1 Prediction Results

The EPL prediction matches are specified clearly on Table 2, highlighting both actual result and
prediction results

Table 2: English Premier League (EPL) Prediction Matches

Home Away Team Actual Prediction

Team Result
Manchester Stoke Manchester Manchester
Chelsea Tottenham Draw Chelsea
Sunderland Man City Man City Man City
Everton Hall City Draw Draw
Arsenal Southampton Arsenal Southampton
Leister City Liverpool Liverpool Liverpool
Swansea QPR Swansea Swansea

5 Conclusion
The ANFIS model was able to make an accurate prediction of 5 out of 7 matches which is promising but
this implies that there might be more factors which were not included in our model that determine the
outcome of a soccer match. So, further study is encouraged to increase the accuracy of prediction of a
soccer match suing more applicable variables.

REFERENCES

[1] Aditya S. T., Aditya P. and Vikesh K. (2013)” Game ON! Predicting English Premier League
Match Outcomes” retrieved from gameON.com

[2] Ian M. and Phil S (2013), ”Forecasting international soccer match results using bivariate
discrete distributions” Centre for Operational Research and Applied Statistics, Salford Business
School, University of Salford, Salford, Manchester UK.

[3] Jang, R (1991). "Fuzzy Modeling Using Generalized Neural Networks and Kalman Filter
Algorithm". Proceedings of the 9th National Conference on Artificial Intelligence, Anaheim, CA,
USA, pp. 762–767. Retrieved on October 15th 2014 from www.irrolecom/article/00098.pdf

[4] Jang, J. (1993). "ANFIS: adaptive-network-based fuzzy inference system". IEEE Transactions on
Systems, Man and Cybernetics Volume23 issue3. doi:10.1109/21.256541.

[5] Jang, S (1997) “Neuro-Fuzzy and Soft Computing “Prentice Hall, Pp 335–368, ISBN 0-13-
261066-3

[6] Jonathan T. (2014), “2014-15 English Premier League and MLS TV”rtrived online from
http://www.philly.com/philly/blogs/thegoalkeeper/NBC-Sports-2014-Premier-League-TV-
schedule-for-August-through-November.html#sd6Wr3imqE4ehOdH.99

[7] Nauck, D., Klawon F. and R. Kruse, (1997), “Foundations of Neuro-Fuzzy Systems”, J. Wiley &
Sons pages 312

Copyright © Society for Sci ence and Education United Kingdom 39

 Amadin, F. I and Obi, J.C.; English Premier League (EPL) Soccer Matches Prediction using An Adaptive Neuro-Fuzzy
Inference System (ANFIS) for. Transactions on Machine Learning and Artificial Intelligence, Volume 3 No 2 April
(2015); pp: 34-40
[8] Negnevitsky, M., (2002) “Artificial Intelligence: A Guide to Intelligent Systems”, 2d ed. Harlow,
England: Addison Wesley, pp. 90-343.

[9] Tony M. (2014), “20 English Premiership Teams” http://www.businessinsider.com/new-

premier-league-kits-2014-8?op=1

[10] Yue W. M. (2003)” Prediction on Soccer Matches using Multi-Layer Perceptron” ID: 903-051-
7735

URL:http://dx.doi.org/10.14738/tmlai.32.1027 40

View publication stats