Web Application for Support in Basketball Game Analysis

Tomislav Horvat

, Ladislav Havaš

and Vladimir Medved

Department of Electrical Engineering, University North, 104. brigade 3, Varaždin, Croatia

Faculty of Kinesiology, University Zagreb, Zagreb, Croatia

Keywords: AssistantCoach, Basketball, Database Analysis, Expert System, ER Model, Web Application.

Abstract: The development of ICT technologies and usage in various fields enable faster development and analysis of

large amounts of data, but also a way of saving the data. Stored data are easier to search, edit and analyse.

Stored data, usually stored in databases, can be analysed and useful information can be drawn. As Internet

communication becomes the dominant way in exchange of data and information, many application and

expert system were developed. In this paper is shown AssistantCoach expert system, which was developed

for supporting basketball coaches in making decisions and analysing games. It describes a new algorithm

for game win percentage prediction and allows users to make notes about opponent players and teams, but

also notes about their own players, trainings and different types of game analysis. Based on input

parameters (IPE in selected time period, predicted parameters) the expert system calculates game win

percentage. By increasing or decreasing parameters, coaches are able to find the highest percentage for the

win and to use time before game to correct team mistakes. In addition to the algorithm for game win

percentage, the expert system offers their user a number of different analysis and statistics. Expert system

uses every input game boxscore and other notes as a new material to learn. Described methods and

algorithm are verified on a selected sample of basketball games and selected time period.

1 INTRODUCTION

Nowadays, the usage of computer analysis in sports,

especially in professional, has become a normal part

of the game analysis (Havaš et al., 2014). Statistical

analysis, also known as notational analysis, has

become interesting not only to coaches and sports

workers, but also to sports fans and reporters

(Hughes and Franks, 1997; Hughes and Franks,

2004). Mathematical presentation of a player's

performance can give a realistic image of player

performance and supports coaches to analyse

player's performance in a quick and easy way, very

often during a game or practice. The goal of the

game analysis is not to establish how good or bad

team or individual performance is, but to find and

correct common mistakes that will lead to increasing

of the individual performance and thus the team

performance. In every sport, especially team sports

like basketball, in which there is an individual or

team approach, using information and

communication technologies (ICT) enabled user's

(coaches, sport workers or sport fans) possibility to

analyse game on a different level (Dežman et al.,

2001). In a situation where there is an expert system

that has the ability to analyse player or team

performance and to extract useful information,

Internet communication becomes dominant among

users and the expert system (Havaš et al., 2013;

Havaš et al., 2013).

In this paper is shown one such expert system

that supports basketball coaches to analyse

individual or team performance in a game by using

web based application (Jackson, 1998; Leondes,

2002). The application can analyse team or player

performance before, during or after the game in real

time and supports coaches in making decisions.

Input data into the expert system’s database are

boxscores of played games, player notes, training

information and opponent notes.

The second part of this paper describes the expert

system architecture while the third part of the paper

shows the development of the expert system through

several stages of development. In the fourth part is

described and shown verification of results for a

selected time period. The final part of the paper

gives a conclusion and the list of references.

Horvat, T., Havaš, L. and Medved, V..

Web Application for Support in Basketball Game Analysis.

In Proceedings of the 3rd Inter national Congress on Sport Sciences Research and Technology Support (icSPORTS 2015), pages 225-231

ISBN: 978-989-758-159-5

225

2 AssistantCoach EXPERT

SYSTEM REPRESENTATION

The prototype of the expert system was developed

by using open-source programs (PHP and MySQL).

The whole application solution was programmed in

first object-oriented version of the language PHP,

version 5.4 (PHP, 2015). The database was designed

and implemented in MySQL relation database,

version 5.1.61 (MySQL, 2015).

2.1 The Architecture of an Expert

System

The first display which users of the prototype of

AssistantCoach application encounter is shown in

the Figure 1.

Figure 1: Opening display of AssistantCoach application

(log in the page).

Users can log in to the application as an

administrator, moderator or guest. The privilege to

add new users has only the administrator. The

moderator has limited privileges and can add new

games, boxscores, notes and can update information

about players. The guest can only see a schedule of

playing, boxscore of every game and player

individual average statistics. When the user has

successfully logged in, the user is presented with a

home page, in accordance with the authorization

level. A specific part of the application, functions or

menu items, is shown or hidden from the user. Home

page, which is shown after successful log in, gives

user the possibility to adjust the time period for

setting start and end date of every analysis.

AssistantCoach application consists of nine menu

items, including Home page, contained in the

navigation bar.

Menu item Players allows application users to

add, update or delete players, adding notes about

players, but also tracking players’ technical and

physical development.

Menu Games gives the user the possibility to add

new games, change the game result and to add

individual player performance during the game.

Menu Games also gives user the possibility to

analyse the game in order to compare with other

games in selected time period, but also a comparison

of players in selected time period. Very interesting

part of Games menu is the game prediction. Expert

system, based on user input parameters (predicted

game parameters and player’s minutes in play)

calculates the possibility for the win. Predicted

parameters are selected based on coach’s prediction,

based on opponent quality and current condition of

his team. Probability to win is calculated according

to the average player and team statistics of the

selected time period and the games against the

selected opponent. Menu Games also gives user the

possibility to make live statistics during the game

and possibility to compare current team and player

statistics to selected time period statistics. Statistics

can be analysed during the game.

Menu Statistics gives user all kinds of statistics

including IPE calculation which will be discussed in

later chapters. Menu Statistics also gives user the

possibility to define and save their own statistics,

according to their own view on the particular game

parameters.

Menu Opponents gives user the possibility to

make notes about opponent teams and players, while

Menu Reports gives user the possibility to make a

report about player in selected time period (average

numbers, notes, individual performance on every

game and training arrivals).

Menu Training gives user the possibility to make

records about every training and intensity of every

aspect of training, but also records about players and

their performance.

Menu About gives information about team staff

while Menu Fan Area behaves as a forum where

fans or reporters can leave their observations about

teams.

While AssistantCoach is Web based application

and authorization is needed, application

AssistantCoach can also behave as club Internet

page. Authorization level allows an administrator to

define the privileges of each user, thus the amount of

displayed information.

2.2 Input Data into Expert System

Input data into the expert system are game

boxscores, game result by minutes, notes about own

players, opponent players, opponent teams and

training information. Boxscore and result change by

icSPORTS 2015 - International Congress on Sport Sciences Research and Technology Support

226

minutes can be inserted into the database using form

Livestats or manually after the game, while all kind

of notes must be inserted manually. Segment of

livestats form is shown in the Figure 2.

Figure 2: Livestats form.

Expert system users can also make notes about

opponent teams or players, but also notes about

trainings and their own players. Notes can support

the coach in making decisions and can be used as a

reminder for previous played games or opponent

players. Flow chart of the expert system is shown in

the Figure 3.

Figure 3: Expert system flow chart.

2.3 The Index of Player Efficiency

(IPE)

Player efficiency index is a statistical data which

mathematically differently evaluates the basic

elements of the basketball game and thus

numerically evaluates the usefulness of a player in a

particular game or time period (Lidora and Arnon,

2000). Statistical analysis, like IPE, is a kind of

notational analysis. Notational analysis is basically

the gathering and analysing of data that has been

obtained by observing the performance of a team or

an individual in a competitive situation (Hoofler and

Payne, 1997; Karapidis et al., 2001). The staff

makes a list of measurable performance indicators as

a list of key elements of a match and then, based on

experience, the team or individual performance can

be evaluated mathematically using the appropriate

algorithm. One such algorithm is Index of Player

Efficiency (IPE).

The main advantage of IPE compared to other

known player efficiency indexes is the fact that IPE

makes players’ defensive activities equal to

attacking activities. Due to the fact that the average

basketball game viewer favours player's attacking

activities (dunks, buzzer beater shots, clutch

shots…) and basketball staff, especially coaches,

analysis of the whole player game, IPE makes

balance between offensive and defensive activities,

especially valuing above average contributions of an

individual in relation to the team. IPE algorithm

consists of a number of components where every

component represents a segment of the game

(points, rebounds, field goal percentage, steals,

turnovers, assists, fouls…). Every component of the

algorithm is then multiplied by particular factor.

Application AssistantCoach allows their users to

adjust the factors in the IPE algorithm according to

their wishes.

2.4 Game Prediction Algorithm

Various analysis and performance indicators have

allowed coaches to make faster and more efficient

analysis of a team or individual. Based on statistics

of previous games and time period selected by the

user, the system can draw useful information and

predict the percentage for the win. Coaches can

predict game parameters of scheduled game and

expert system will, based on statistics of previous

games, calculate percentage for the win. By

increasing or decreasing some parameters, coaches

are able to find the highest percentage for the win

and use time before game to correct team or

individual mistakes during the game.

Using decision trees application AssistantCoach

can predict the percentage for the win based on

predicted parameters and statistics of games played

in the selected time period. Decision tree learning is

a method usually used in data mining with a goal to

create a model that predicts the value of a target

Web Application for Support in Basketball Game Analysis

227

variable based on input data. Input data into

AssistantCoach application are boxscores of played

games.

The Application AssistantCoach uses decision

tree for every parameter of the game. The output of

one decision tree is input into another decision tree.

The initial percentage of victory is set to 50%.

Application AssistantCoach gives user the

possibility to set parameters according to users’

prediction. Parameters of prediction can be divided

into three groups:

 predicted team parameters – predicted group

statistics of a particular game (team field goal

percentage, offensive and defensive rebounds,

steals, assists and turnovers)

 individual parameters – predicted time in play

for every player

 opponent team parameters – opponent team

name, venue, predicted opponent team quality in

relation to users’ team and number of opponent

team offensive rebounds

The application compares team average statistics in

the selected time period to users’ predicted numbers

and based on result increases or decreases the

percentage for the win. If some of the parameters

remain empty, the comparison for that particular

parameter is neglected. Predicted game form is

shown in the Figure 4.

Figure 4: Predict game form.

As it is shown in Figure 4, user predicts game

parameters, predicts minutes in play for every player

and predicts and defines opponent performance and

quality. Some parameters can remain empty or

unpredicted, a decision tree for the particular

parameter is skipped and the percentage for the win

remains unchanged. Basic game parameters (Game

parameters in the Figure 4) are calculated based on

average team performance in the selected time

period. Worse predicted parameters decrease

percentage for the win while better predicted

parameters increase percentage for the win. Based

on the ratio of predicted and average parameters, the

expert system defines coefficient, which is

multiplied by the current percentage for the win.

Current percentage for the win is output from

previous decision tree. The expert system also

analyses similar game parameters and based on

average and predicted performance also defines the

coefficient. As noted earlier, empty parameters do

not change the percentage and every parameter

makes one decision tree. Output of the decision tree

is input into another decision tree. Also, as noted

earlier, input into first decision tree is coefficient 0.5

or 50%. Prediction of more parameters and selecting

a longer time period, the expert system can more

accurately calculate the percentage for the win.

The second group of parameters, Players in the

game in the Figure 4, predicts minutes in play for

every player. The expert system, based on average

IPE which is introduced in the Chapter 2.3 the Index

of Player Efficiency (IPE), calculates coefficient

which is multiplied by the current percentage for the

win.

Last group of parameters, Opponents parameters

in the Figure 4, defines and predicts the opponent

quality and venue. The expert system, based on

results in the selected time period against the same

and other opponents, predicts opponent offensive

rebounds, opponent quality and venue calculate

coefficient which is then multiplied by the current

percentage for the win. The Figure 5 shows the

example of a decision tree where the coefficient for

assists is calculated. Based on the ratio between

predicted assists number and the average team

assists per game, the coefficient of the particular

ratio is multiplied with current win percentage

(w_perc). The output of assists decision tree is input

into steals decision tree. The decision tree is built for

every component. If predicted parameter remains

empty, the win percentage remains unchanged.

predictedassists/averageteamassists

coef=1,06 coef=1,03 coef=1 coef=0,97 coef=0,94

<1.3,

+∞>

<1.1,1.3] [0.9,1.1]

[0.7,0.9

<‐∞

,0.7

w_perc=w_perc+(coef–1)*w_perc

predictedsteals/averageteamsteals

Figure 5: Decision tree for assists.

icSPORTS 2015 - International Congress on Sport Sciences Research and Technology Support

228

3 DEVELOPMENT OF THE

EXPERT SYSTEM

The development of an information system passes

through several stages of development; conceptual

design phase, implementation phase and phase of the

physical design. Conceptual design, also known as a

conceptual model, identifies the highest-level

relationships between different entities. Conceptual

model of the data is usually represented by ER

(Entity - Relationship) model which is a graphical

representation of entities and relationships among

them, typically used in computer sciences to present

the organization of data within databases and

information systems (Chen, 1976). ER model of the

expert system is given in the Figure 6.

GAME

PLAYER

plays

OPP_TEAM

plays

NOTES

TECHNICAL

_NOTES

is_for

OPP_TEAM

_NOTES

OPP_PLAYERS

is_for

OPP_PLAYER

_NOTES

plays

is_for

PHYSICAL

_NOTES

TRAINING attendts

Figure 6: ER model of the expert system.

ER model in the Figure 6 shows the relationships

between entities in AssistantCoach application on

conceptual design phase. ER model says that the

game is played by more players and the player plays

more games. Every note, regardless if it is made for

a player or an opponent team is for one player or

team and player or opponent team can have more

notes. ER model in the Figure 6 also says that the

player attends more trainings and the training is

attended by more players (relationship between

entities TRAINING and PLAYER). Relation

OPP_PLAYERS – OPP_TEAM says that the player

plays for one team and in a team plays more players.

Creating an ER model completes the first phase of

the development of expert system. The next step in

the development of an expert system is the creation

of a logical model. The resulting ER model is

translated into a relational schema. Steps and

conversion rules for converting into relational model

are known from earlier. Relational model is then

implemented into the database management system.

The relational model of an expert system is shown in

the Figure 7.

OPP_PLAYER_NOTES

OPP_PLAYERS

OPP_TEAM

OPP_TEAM_NOTES

GAME PLAYERSTATISTICS

TRAINING

PLAYER_TRAINING

PHYSICAL_NOTES

TECHNICAL_NOTES

NOTES

IDPK

date

descripti

player_id

nam e

surname

bir th_date

opp_tea m_

IDPK

name

IDPK

date

des cription

opp_tea m_

host_idFK

guest_idFK

date

result

name

surname

birth_dat

height

weight

player_id

FKPK

min

points

......

game_id

FKPK

date

descripti

player_id

FKPK

training_

FKPK

date

descripti

player_id

date

des criptio

pla yer_id

date

descripti

player_id

Figure 7: Relational model of the expert system.

Expert system, Web based application

AssistantCoach, was developed with the purpose of

enabling the exchange of information in real time.

The application works on client-server architecture.

Client (user) sends a request to the server, which

passively waits, receives the request and sends a

response back to the client. The client can connect to

the Web page using a variety of devices that have

access to the Internet. The client sends a HTTP

request by writing the Web address into a Web

browser and Web server sends the answer back to

client. Architecture of Internet communication of

AssistantCoach application is shown in the Figure 8.

The developed expert system has been verified in

the selected time period. Games of the season

2014/2015 in Croatian 2nd basketball league were

analysed. For every game, team and individual

statistics were made, inserted into the database and

analysed by the developed expert system.

Web Application for Support in Basketball Game Analysis

229

(1)DNSrequest

(3)HTTPrequest

(4)Databaserequest

(2)DN Sreply

(5)Databa sere ply

(6)HTTPreply

Figure 8: Internet communication architecture.

Predictions were made based on average parameter

before the game was played, real parameters and no

parameter, just based on the team opponent name

and venue. The time period used for predictions was

whole 2014/2015, including friendly, Cup and

league games. Predictions were calculated for the

time period up to a match, while prediction based on

real parameters was calculated after the game was

played, also for time period up to a match. Table 1

shows game prediction of a league game of season

2014/2015. Columns colored in red show wrong

prediction.

4 RESULTS

As was mentioned earlier, Table 1 shows results

based on game prediction results during whole

2014/2015 season. Table 1 consists of five columns;

first column shows game date, the second column

shows the game result where prefix L means lost

game and prefix W means won game, the third

column shows win percentage based on average

parameters up to the game, the fourth column shows

game win percentage based on real game parameter

up to the game and the fifth column shows game win

percentage based only on opponent name and venue.

Fifth column gives prediction based only on

previous game against the same opponent and ratio

of win and lost games based on venue in selected

time period, so it is called “no parameter

prediction”. As is evident, time period and number

of played games significantly affect game prediction

algorithm. A greater amount of data in the database

reduces the possibility of extreme data which will be

included in the analysis.

Extreme data usually appear in the case of a

small sample data when the expert system does not

have enough data to learn and to extract useful

information. In the case of larger amounts of data

occurrence of extreme data is significantly decrease,

and thus analysis and game prediction percentage

more accurately. Based on the results from Table 1

and a sample of 21 matches and 63 predictions, it is

Table 1: Game prediction results.

DATE RESULT

AVERAGE

PARAMETER

PREDICTION

REAL PARAMETER

PREDICTION

NO PARAMETER

PREDICTION

11.10.2014. L 69:75 47% 42% 54%

18.10.2014. L 70:69 60% 43% 46%

26.10.2014. W 97:70 81% 71% 54%

2.11.2014. W 64:78 83% 79% 60%

8.11.2014. W 91:68 95% 49% 84%

15.11.2014. W 82:101 100% 100% 87%

22.11.2014. W 99:79 67% 78% 67%

29.11.2014. W 85:120 74% 100% 83%

6.12.2014. W 100:82 59% 91% 59%

14.12.2014. W 98:111 94% 85% 83%

19.12.2014. W 112:65 99% 100% 84%

11.1.2015. W 80:87 50% 59% 50%

18.1.2015. L 79:84 63% 53% 70%

31.1.2015. W 96:75 68% 75% 75%

7.2.2015. W 64:111 86% 94% 98%

14.2.2015. W 95:64 69% 67% 79%

21.2.2015. W 82:90 54% 81% 62%

28.2.2015. W 117:97 73% 100% 75%

8.3.2015. L 81:75 60% 45% 62%

13.3.2015. W 114:109 73% 91% 75%

22.3.2015. W 67:108 96% 100% 98%

icSPORTS 2015 - International Congress on Sport Sciences Research and Technology Support

230

clear that the game prediction algorithm was wrong

on 8 predictions. The biggest mistake is recorded for

the game played on 18.1.2015. The reason for this is

the fact that the match was played without five very

important players, players with the highest average

IPE.

5 CONCLUSIONS

This paper shows modelling of a Web based expert

system for support in basketball training and game

decisions. Expert system allows their user to easily

analyse basketball games or make good preparation

for upcoming games based on previous analysed

games or notes made about opponent teams or

players, but also previous played games. This paper

shows complete development of the expert system.

Game analysis is very complex matter. It takes years

of experience and knowledge to analyse game,

especially if it is made by hand. Developed expert

system supports coaches to analyse game on fast and

efficient way.

Second chapter presents expert system

architecture and shows expert system flow chart.

Input data into the database are boxscore and notes

about own players, opponent teams and opponent

players. Very interesting information about player

performance are player efficiency indexes. The

developed expert system uses IPE (Index of Player

Efficiency) index, statistical data which

mathematically differently evaluate basic elements

of basketball game and thus numerically evaluates

the usefulness of a player. The main advantage of

IPE compared to other known player efficiency

indexes is the fact that IPE makes players’ defensive

activities equal to attacking activities.

Expert system, based on IPE, previously played

games and predicted parameters, calculates game

win percentage prediction. By changing parameters

coaches are able to find the highest percentage for

the win and use time before game to correct team

mistakes. The initial percentage of victory is set to

50%. The Application AssistantCoach uses decision

tree for every parameter of the game, by increasing

or decreasing win percentage based on ration

between average and predicted parameter. The

output of one decision tree is input into another

decision tree. Parameters of prediction can be

divided into three groups; predicted team

parameters, individual player parameters and basic

opponent team parameters.

Third chapter presents the development of the

expert system through phases of conceptual and

logical phase. Conceptual model of the data is

represented by ER (Entity - Relationship) model

which is a graphical representation of entities and

relationships among them. The resulting ER model

is translated into a relational schema, which is

implemented into MySQL database. The application

works on client-server architecture.

Verification of the algorithms and the methods

has been conducted on a basketball 2014/2015

season.

REFERENCES

Chen, P., 1976. The Entity-Relationship Model -- Toward

a Unified View of Data. ACM Transactions on

Database Systems.

Dežman, B., Trninić, S., Dizdar, D., 2001. Expert model of

decision-making system for efficient orientation of

basketball players to positions and roles in the game-

empirical verification. Collegium Antropologicum.

Havaš, L., Medved, V., Skočir, Z., 2013. Application of

mobile technologies in the preparations for long

distance running. icSPORTS 2013.

Havaš, L., Medved, V., Skočir, Z., 2014. Machine

learning for individualized training support in

marathon running. icSPORTS 2014.

Havaš, L., Skočir, Z., Medved, V., 2013. Modeling of the

athlete's training decision support. Technical Gazette.

Hoofler, R., Payne, J., 1997. Measuring efficiency in the

national basket association. Econ lett.

Hughes, M., Franks, I.M., 2004. Notational Analysis of

Sport Systems for better coaching and performance in

sport. London: Routledge.

Hughes, M. D., Franks, I.M., 1997. Notational Analysis of

Sport. London: E. & F.N. Spon.

Hughes, M. D., Franks, I.M., 2004. Notational Analysis of

Sport 2nd Edition – better systems for improving

coaching and performance. London: E. & F.N. Spon.

Jackson, P., 1998. Introduction to Expert Systems (3 ed.).

Addison Wesley.

Karapidis, A, Fotinakis, A, Taxildaris, K., Fatouros, J.,

2001. Factors characterizing a successful

performance in basketball. J Hum Mov Stu.

Leondes, C. T., 2002. Expert systems: the technology of

knowledge management and decision making for the

21st century. Academic Press.

Lidor, R., Arnon, M., 2000. Developing indexes of

efficiency in basketball: talk with the coaches in their

own language. Kinesiology.

MySQL. Retrieved on April 15 2015 from

https://www.mysql.com/.

PHP. Retrieved on April 15 2015 from http://php.net/.

Web Application for Support in Basketball Game Analysis

231