A Graded Concept of an Information Model for Evaluating

Performance in Team Handball

Friedemann Schwenkreis

DHBW Stuttgart, Paulinenstr. 50, 70565 Stuttgart, Germany

Keywords: Information Model, Effectiveness Index, Team Handball, Performance Indicator.

Abstract: Although team handball is a very popular sport in Europe, computer science did almost completely ignore

that area in the past. This article introduces a graded approach for an information model that allows to express

the effectiveness of a team as well as of single players, thus providing a basis for information-based decisions

of coaches, as well as for applying analytical methods. From this perspective, the article is an early step to

further introduce digitalization via a data model into a very classical sports area, however, introducing

mechanisms that take into account the available degree of digitalization.

1 INTRODUCTION

Team handball is a very fast game with full physical

contact of the players who are not wearing any

protectors. The goal rate easily exceeds one per

minute. That is one of the reasons why it has become

a pretty popular game in Europe. In Germany about

750.000 people are playing handball in multiple

leagues (DHB, 2019). The first German league

(Handball Bundesliga) has an annual budget of

approximately 80 million Euros. The top leagues in

France and Spain even exceed that number. Top

teams in France for instance, have an annual budget

between 17 and 18 million Euros (DPA, 2016).

Although we can recognize an increasing usage of

computer science methods from analytics and Big

Data (Morgulev et al, 2018), this is usually restricted

to the premier leagues of sports areas with huge

annual budgets—far beyond the previously

mentioned figure. Hence, team handball is just at the

starting point of digitalization. So far, there is almost

no usage of sensors to automatically collect data of

the players—particularly not during a game. Even the

official game reporting of major leagues just recently

switched to an online platform as a first step into

digitalization (DKB HBL, 2019). Thus, third party

providers are offering a service to collect and provide

game information based on the manual collection of

data, called scouting (Sportradar, 2015).

https://orcid.org/0000-0003-4072-0582

Since video based collection of information like

in football (ChyronHego, 2016) is too expensive and

sensor based information collection (Kinexon, 2017)

is not yet applicable due to the size and complexity of

the equipment, automated information collection is

rather rare in case of team handball. Hence, there are

only some cases for which information of real

handball games have been collected. Even worse,

with the absence of collected information there is also

an absence of insights based on the collected

information. As a consequence, so-called player

effectiveness indexes (PEI - also called player value

indexes) are only at their beginning for team handball.

Currently, there are almost only trivial data models

that allow to evaluate the player performance and

only some advances have been made just recently

(e.g. (Wagner, 2014)). Thus, team performance is

mostly evaluated based on the experience of coaches

rather than on objective, information based

indicators.

Furthermore, due to the fact that the effectiveness

indicators are missing, young potentials are hard to

identify and the development of potentials is very

hard to track. Currently, this area is also dominated

by the intuition of coaches and scouts.

196

Schwenkreis, F.

A Graded Concept of an Information Model for Evaluating Performance in Team Handball.

DOI: 10.5220/0007920001960202

In Proceedings of the 8th International Conference on Data Science, Technology and Applications (DATA 2019), pages 196-202

ISBN: 978-989-758-377-3

2 BASICS OF THE MODEL

Before defining an actual effectiveness indicator

(which is just a domain specific performance

indicator), it is crucial to understand the most

important principles of the domain area — team

handball in our case (Brand, 2008). We can start with

a few simple observations. The overall objective of a

team in a game is to win the game by scoring more

goals than the opponent team. Hence,

 scoring goals is considered to be positive.

 any activity to help scoring a goal for the own

team is considered to be positive. These activities

are called offense activities.

 receiving goals is considered to be negative.

 any activity preventing the opponent team to score

a goal is positive. These activities are called

defence activities.

 losing possession of the ball is negative because a

goal can only be scored when possessing the ball.

2.1 Direct Offense Activities

To actually score a goal, it is necessary to possess the

ball and to throw the ball at the goal (also named

shot). This is called a scoring attempt. Three

outcomes are differentiated when a scoring attempt

occurs: success, miss, and a defended shot. Besides

throwing the ball at the goal, players can also pass the

ball to another player and errors can occur while

passing balls, leading to a loss of the ball: bad pass.

2.2 Direct Defence Activities

The major objective of the defence is to prevent the

opponent team from scoring goals as well as getting

the ball. The following events are differentiated in

case of the defence: steal, block, and save (goal

keeper). All events that lead to the possession of the

ball are additionally categorized as turnover events.

2.3 Rule Violations

In general, it is allowed to interfere with an opponent.

The referees decide whether an interference was

according to the rules (no whistle), or whether there

was a violation of the rules: foul (IHF, 2018). In case

of a violation of the rules there is a whistle and the

violating team loses the ball (aka technical error).

Furthermore, depending on the severity of a foul there

can be an additional sanction (and combinations

thereof).

2.4 Player Position and Movement

Players can move arbitrarily on the field, but they are

neither allowed to run over a non-moving opponent

player nor to step on the penalty area (except for goal

keepers). Overall, in our information model every

player has an associated 5-tuple (n, l, w, d, s) which

describes the players number n, the current position,

expressed as two coordinates l and w, the direction d,

and the speed s of the player. The position and

movement of the ball are modelled with a similar

approach, having the number -1: (-1, l, w, d, s) and d

set to the movement direction of the ball.

In team handball, players can be substituted at any

point in time. However, there are some rules

regarding the substitution procedure. Violating the

rules results in a sanction of the violating player.

Having the objective to define a player effectiveness

index, we need to distinguish players who are

currently playing from players on the substitution

bench. If players are currently on the match field, they

are called active If they are currently on the

substitution bench, they are called inactive.

3 EFFECTIVENESS INDEXES

Given the model elements of the previous section

Basics of the Model, effectiveness indexes can be

defined based on these model elements. However,

there needs to be a detection of the model elements

themselves as well as of combinations (composites)

in order to compute an effectiveness index that is

based on them. The following sections will describe

the options based on the ability to detect the events.

3.1 Lowest Level Index

Even the lowest level leagues of team handball in

Germany have to keep track of a set of basic events

during a game in order to prepare an official game

report that allows to prove the outcome of a match.

The following data are recorded:

 Team players: no distinction of active and inactive

 Goal events: who scored when

 Sanctions of players: who was sanctioned and

when

There is no recording of further events, positions

or movements. Thus, only a very simple index can be

derived from this information just based on the fact

that scoring goals is positive and being sanctioned is

negative. A player can earn a certain number of

A Graded Concept of an Information Model for Evaluating Performance in Team Handball

197

credits by scoring a goal and a player can lose credits

by being sanctioned. Hence, this is not a very useful

index. However, in many cases we do not have more

information than just the official match report.

Figure 1: Zones to record a scoring attempt.

3.2 Single Player, Outcome-based

Index

The next level of detail of tracking events of team

handball was introduced by companies collecting

information for betting (e.g. (Sportradar, 2015). Since

betting service providers are getting payed for

offering a continuous information stream regarding

team handball matches, they are paying human

observers, so-called scouts, to record events during a

game.

The event recording is done manually using a

computer-based tool that sends the events online to a

platform which is then used by the betting companies

(DKB HBL, 2019). The concept has been adapted by

the upper leagues in Germany like the Bundesliga as

well as by the European and the International

Handball Federation for international matches.

Since humans can only record a limited set of

events, basically, only events are recorded that either

result in a goal or in a turnover. The position of an

attempt to score a goal is also recorded but only based

on 8 sectors (see figure 1). Therefore, the following

information is available in addition to the approach of

the lowest level index, if the event results in a

turnover (and ends the current attack):

 Miss events: shooter with sector and time

 Steal events: defender and time

 Block events: shooter with sector blocking player

and time

 Save events: shooter with sector, goal keeper and

time

This introduces at least some information

regarding defence players and goal keepers. Thus,

allowing to earn credits for offense actions like

scoring but also for defence actions like saves and

blocks (Menz, 2017). In addition, there are

approaches for indexes introducing penalties for non-

successful scoring attempts as well as the concept that

the amount of credits gained or lost depends on the

sector from which the scoring attempt was made

(Thiele, 2017). Due to the ability to differentiate

offense from defence actions, we can also introduce

two separate indexes in which we aggregate the

corresponding credits: an offense index and a defence

index. The overall effectiveness index is then the

aggregation of the two.

Unfortunately, there is still no information

recorded that allows to determine which players are

active at a certain point in time. Although only the

active players of a team should be able to earn credits

when a goal is scored or prevented, the collected

information does not allow that distinction. Thus,

current approaches focus only on the players directly

involved when an event occurs.

The limitations of the indexes in this section are

again due to the limited information that is manually

collected by the scouts. This is mainly caused by the

limited capacity of humans to record the events. The

available capacity is currently spent to collect

information symmetrically for both teams of a match.

If that approach is changed to mainly focus on one

team only (for each team) an enhanced index

becomes possible. For instance, it can be recorded

which player of the own team was involved when an

opponent player was sanctioned. Thus, adding credits

to the offense index of the player who was the fouled

because sanctioning of an opponent player is

beneficial for the own team.

3.3 Active Player, Outcome-based

Index

Neither scoring goals nor preventing goals in team

handball can be usually achieved by a single player.

Thus, all active players contribute and should receive

credits accordingly. I.e., as soon as we can detect who

is active when an event occurs, we can also add

credits to the index of the rest of the active team

members and not only to the directly involved player,

based on an assumed degree of contribution (Thiele,

2017) and (Uhrmeister et al, 2019).

A direct consequence of being able to distinguish

between active and inactive players on the level of the

effectiveness indexes is, that the index of currently

active players can be compared with their own

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

198

indexes of the past (expected index value) as well as

with historical indexes of currently inactive players,

which can be used as an information to base

substitution decisions on (regarding which active

player to substitute by which inactive player).

However, keeping track of player substitutions in

team handball is very exhaustive and needs two

additional scouts if done manually. Since this is much

too costly, we are in the process of evaluating a

cheaper sensor-based approach to detect a

substitution automatically, which is then just used as

additional input that is merged with the asymmetric

scouting information. Thus, allowing to detect which

player becomes active and which player becomes

inactive.

3.4 Active Player, Contribution-based

Index

The previously described indexes did not take any

individual contribution into account except for the

activity of the finally involved player. Since team

handball is a team sport, these approaches ignore a

crucial aspect of a team sport: the team coordination

of the activities of players. That means that there

might be very valuable players without being directly

involved in the recorded events.

The reason why this information is not

considered, lies in the fact, that it is not decidable if a

certain activity of a player will be connected to a

future scoring attempt or whether it will prevent a

future goal respectively. Only by the concept of

backtracking after a scoring attempt (or turnover

without a scoring event) it can be determined whether

there was a direct connection or not. Furthermore, it

is necessary to observe the activities of all players in

parallel, which can only be achieved using modern

sensor technology.

In case of the contribution-based index introduced

in this section, we assume that we can track the

position and movements of players (and the ball) as

described in section 2. using current sensor

technology with modern computer equipment, as in

case of IoT scenarios. There are multiple vendors

offering solutions in these areas.

3.4.1 Potential Scoring Probability

If we recall the discrete positions of a field depicted

in Figure 1, then we can associate a scoring

probability with each of the cells (see Figure 2). The

initial scoring probability is the probability to score a

goal when no defensive player except the goal keeper

can interfere. The defence players can reduce this

probability by getting in between the shooter and the

goal and even further by getting in physical contact

with the shooter (see also section 3.4.3).

Figure 2: Potential scoring probability.

In case of the contribution-based index, we

assume that the position of players can be detected,

and a potential scoring probability can be computed.

For a further advanced model, a specific potential

scoring probability distribution can be defined for

each player. Thus, allowing to consider the different

player abilities. For instance, a wing player usually

has a high probability from the sides of the field but a

relatively low probability from the central back (see

Figure 3).

Figure 3: Player specific scoring probability.

3.4.2 Player Move Segments, Team Moves

and Team Tactics

To have just the raw positions and movements of

players does not actually allow to determine the value

of the movement activity. For that purpose, we need

A Graded Concept of an Information Model for Evaluating Performance in Team Handball

199

to combine the movements to segments which consist

of a starting cell and an end cell, meaning the player

has moved from the starting cell to the end cell. The

segments are usually characterized by the fact that the

player starts moving at the starting cell and keeps the

general direction and speed until he or she reaches the

end cell. I.e. a significant change of the speed or the

direction of a player ends the current segment and

starts a new one.

Particularly offense tactics combine the segments

of multiple players and passes to an overall team

move that usually even has a specific name (e.g.

“Sperre-Absetzen” which is similar to “pick and roll”

of basketball) and is explicitly trained (Brand, 2008).

Key is, that a team move is a certain, in general

parallel, combination of segments of multiple players.

The set of moves which a team can perform is also

called the team tactics. There are team tactics for

offense as well as for defence. We assume that

movement segments of players can be detected by a

relatively simple interpretation of a player’s position

and movement information. Furthermore, we assume

that we can even detect passes, which depends on the

used sensor technology and the used ball respectively.

Team moves are expected to be detectable soon by

using analytics and deep learning mechanisms based

on video streams.

Figure 4: Pressure situations.

3.4.3 Concept of Pressure and Its Prevention

The concept of pressure in team handball is defined

based on the potential scoring probability of a

position. Pressure has the objective to reduce the

scoring probability of a position when an offense

player is located at that position. A defence player can

reduce this probability by moving in the area with a

direct line between the position and the goal (see

Figure 4). Multiple defence players can further reduce

the scoring probability by combining their efforts by

a team block or defensive wall (in case of a free

throw) to cover the complete area.

The scoring probability of a position is similarly

reduced by the physical contact of a defence player

with an offense player being at the position, while the

defence player is not in a line between the position

and the goal (see Figure 4). I.e. the defence player has

a distance of up to two field cells to the position. The

highest reduction of the probability by a single

defence player is caused by being in direct contact

distance while standing in the line to the goal.

To prevent the pressure by a defence player,

offense players hinder the defence player to move in

the line between a position and the goal and from

getting close to a position respectively, by blocking

the way. I.e. there is no single move segment that

allows the defence player to interfere without running

over the preventing player.

3.4.4 Assessment of Offense Activities

The basic idea of the contribution-based index is that

players earn credits when even contributing only

indirectly to a scoring attempt. This means, that in

addition to the credits earned in case of the outcome-

based indexes, further credits can be earned. The

following activities of players lead to additional

credits (presumed that the activities can be detected):

 Passing the ball to the player who attempts to

score if the ball receiving player had a position

with a higher potential scoring probability

(sometimes called assist)

 Having a position on the field which prevents an

opponent player from causing pressure (barrier)

on the attempting player.

 Having a last movement segment that increased

the scoring probability of the position of the

player, thus increasing the threat on the other

team.

 Participation in the last detected team move of the

same attack that ended in the scoring attempt

(tactical participation).

 Being the fouled player if the defence player is

sanctioned.

An offense foul and technical errors lead to a loss

of credits.

3.4.5 Assessment of Defence Activities

Players can also earn credits in addition to the direct

defence activity related credits (steal, block, and save)

by positioning themselves appropriately when a

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

200

scoring attempt by the opponent team is made.

Players earn defence credits for:

Having a position which decreased the scoring

probability of another attacking player. The more the

scoring probability is reduced, the more credits are

earned (threat reduction).

 Preventing a goal after a team move of the offense

team. (tactics response)

 Interrupting an attack by a foul or getting the ball

out of play without a sanction (interrupt)

 Being in the direct line between the attempting

player and another offense player with a higher

scoring probability (pass prevention)

Being sanctioned leads to a loss of credits.

3.4.6 Overall Index Calculation

Whenever there is a turnover or a scoring attempt, the

indexes are updated. For each player, the offense

credits and the defence credits are determined and

then aggregated to the offense index, the defence

index as well as to an overall index which gives an

indication of the effectiveness of players. It is obvious

that inactive players cannot earn credits in case of the

contribution-based index. The credits can even be

aggregated on a team level.

We do not go into details regarding the

aggregation at this point, because we assume that the

calibration (see next section) will reveal criteria for

deciding on the appropriate method for aggregating

the credits. However, we assume that weighting

factors are needed rather than using just a linear

summation.

3.4.7 Calibration

A key aspect of almost all the indexes and particularly

of the contribution-based index is the amount of

credits that can be earned or lost in each of the cases

as well as the used aggregation function. The actual

contribution weight of each activity that leads to

credits heavily depends on the domain of team

handball. Thus, an in-depth calibration phase is

needed to assign the appropriate amount of credits to

the described activities.

By using a first pair-wise comparison, a starting

point for the calibration can be derived:

 The value of scoring a goal is similar to either a

steal, save or a block with a turnover.

 In case of offense, increasing threat should have

the lowest value followed by the barrier and the

assist on the same level. The value of tactical

participation should be similar to the value of

increasing threat and being fouled with a sanction.

 An offense foul or a technical error should result

in a loss equivalent to the value of a barrier.

 Pressure by moving in between the line of an

attacking player and the goal should have the

lowest defence value. The pass prevention should

have a similar value.

 Pressure by contact should have a higher value

than the pass prevention but lower than the save

or a block without a turnover which again have a

lower value than a steal.

 Being sanctioned should result in a loss equivalent

to a block without a turnover.

These rules can be depicted on a simple point

earning system using the straight-forward summation

as the aggregation. Whether this will lead to results

that are like domain experts’ opinions needs to be

determined and the credit model needs to be adapted

accordingly.

4 CONCLUSIONS

This paper has introduced a concept of an information

model that allows to calculate an effectiveness index

for team handball players as well as whole teams

based on earning or losing credits for certain

activities.

The presented approaches heavily depend on the

information collected during a match and it has been

shown that some information cannot be collected by

humans as it is done exclusively today. Thus, the

introduction of digitalization in the context of the

information collection in team handball is very likely

to have a significant impact on future decisions of

team handball coaches not only during games but also

regarding training of players and team management.

To collect the necessary data for the introduced

contribution-based index we are currently

investigating a combination of passive RFID based

technology, as well as active sensor technology and

near-line video stream analysis. Even though we

assume that the automated team move detection in

team handball will probably need a few more years of

development, the structure of the index concept

allows to start without that technology and to include

that part at a later point in time when it becomes

available.

It is important to notice, that the work presented is

at the concept level. Still, several details need to be

worked out. However, the explicit description of the

concept has helped to convince the application area,

the team handball community, that introducing IT

concepts and methods is not just a current hype but

make sense and can actually help to get decisions in

A Graded Concept of an Information Model for Evaluating Performance in Team Handball

201

team handball to a new level. Additionally, the

introduction of the indexes and the collection of the

additional player information allows the application

of analytics in order to find patterns which help to

improve teams.

A side effect of explicitly expressing the value of

activities is the understanding of the game and

allowing to simulate as well as analyse game

situations, which was too complex for human beings

in the past. It is very likely that this will generate new

insights in the future - particularly in the context of

team moves and tactics.

ACKNOWLEDGEMENTS

We would like to thank the team handball coaches

and other handball experts who have contributed to

develop this concept. Namely Eckard Nothdurft

(coach of TSV Neuhausen), Karsten Schäfer (coach

of TVB Stuttgart), and Ralf Bader (coach of SG BBM

Bietigheim).

REFERENCES

Brand H. 2008, Handball. Mein Spiel, mein Stil. Trainieren,

Spielen ,Coachen. Philippka-Sportverlag.

ChyronHego 2016, TRACAB Optical tracking Product

Information Sheet. ChyronHego.

DHB 2019, Unser Markenleitbild. https://dhb.de/der-

dhb/verband.html Accessed March 19, 2019

DKB HBL 2019: Liveticker.http://www.dkb-handball-

bundesliga.de/en/liveticker/dkb-hbl/liveticker/ .

Accessed March 19, 2019

DPA 2016, Start der EHF-Champions League: THW Kiel

und SG Flensburg mssen Vollgas geben.

https://www.shz.de/14920136 Accessed March 19, 2019

IHF 2018, IX. Rules of the Game. http://www.ihf.info/files/

Uploads/NewsAttachments/0_New-Rules of the Game

GB.pdf. Accessed March 19, 2019

Kinexon 2017, Real-time Performance Analytics. Kinexon.

Menz T. 2017, Systematische Spielanalyse im Sportspiel

Handball – Ein Index zur Quantifizierung der

defensiven Spielwirksamkeit sowie Potentiale der

statistischen Datenerfassung. Bachelor Thesis, German

Sport University Cologne.

Morgulev E., Azar O., Lidor R 2018, Sports analytics and

the big-data era. https://link.springer.com/article/10.

1007/s41060-017-0093-7, Accessed March 19, 2019

Sportradar 2015, Handball Scout Admin (HAS) Manual.

Sportradar AG.

Thiele J., 2017, Systematische Spielanalyse im Sportspiel

Handball - Ein Index zur Quantifizierung der offensiven

Spielwirksamkeit sowie Potentiale der statistischen

Datenerfassung, Bachelor Thesis, German Sport

University Cologne.

Uhrmeister J., Brosig O., 2019, Ausgabe eines

Rangfolgeverfahrens zur Handball-WM 2019 – Der

„PlayerScore“, Leistungssport 2/2019, DOSB

Wagner H., Finkenzeller T., Würth S., von Duvillard S.,

2014, Individual and Team Performance in Team-

Handball: A Review, Journal of Sports Science and

Medicine, 2014, 13, pp. 808-816.

DATA 2019 - 8th International Conference on Data Science, Technology and Applications

202