=Paper=
{{Paper
|id=Vol-3341/wm4727
|storemode=property
|title=Providing Training Plans for Weight Training using CBR
|pdfUrl=https://ceur-ws.org/Vol-3341/WM-LWDA_2022_CRC_4727.pdf
|volume=Vol-3341
|authors=Alexander Weber,Pascal Reuss,Jobst-Julius Bartels,Klaus-Dieter Althoff
|dblpUrl=https://dblp.org/rec/conf/lwa/0006RBA22
}}
==Providing Training Plans for Weight Training using CBR==
https://ceur-ws.org/Vol-3341/WM-LWDA_2022_CRC_4727.pdf
Providing Training Plans for Weight Training using
CBR
Alexander Weber1 , Pascal Reuss1,2 , Jobst-Julius Bartels1 and Klaus-Dieter Althoff1,2
1
Intelligent Information Systems Lab, University of Hildesheim
2
Competence Center Case Based Reasoning, German Center for Artificial Intelligence, Kaiserslautern
Abstract
Since several years the interest in fitness is increasing and therefore the demand for information about
training plans has also increased. While fitness plan, especially training plans for weight training, can be
created online with various tools, these plans usually are not tailored to the specific user, but mostly
generic for certain purposes. This paper presents an application that uses Case-based Reasoning (CBR)
to generate and adapt training plans for weight training based on basic user data, preferences, and
restrictions. We describe the idea and goals for the application as well as the knowledge model for the
CBR component including case structure and similarity measures. We also give an short overview of the
current implementation and a first evaluation of the application.
Keywords
Case-based Reasoning, Case-based Planning, Fitness plans, Weight training
1. Introduction
The World Health Organization (WHO) names physical inactivity as one of the major causes for
a decreased lifespan in developed countries on earth. Physical inactivity has negative impact
on several diseases, for example diabetes[1]. The interest in fitness and the number of people
performing physical activities to enhance and maintain their health is increasing during the
last years. As a consequence of this increasing interest in fitness, the fitness industry itself
is growing also. But despite this constant increase, a large number of people still maintain
an nonathletic life style with insufficient physical activity[2]. The last two years with the
COVID-19 pandemic have not supported the positive trending, because many physical activities
like gyms were restricted and in many cases those missing activities are not replaced by other
activities[3].
When someone decides to become active or more active in sports, they have a variety of
digital resources available in different forms. The offerings range from paid models, such as
Peloton, to free resources such as YouTube videos or free apps. Depending on the offer, the user
gets access to personal advice or prefabricated plans, which are made available online. Sport is a
very individual activity, where training plans are ideally tailored to the user and are performed
with a coach to ensure that the exercises are performed correctly, especially in weight training.
Many online resources, such as prefabricated training plans, offer little or no opportunities for
LWDA’22: Lernen, Wissen, Daten, Analysen. October 05–07, 2022, Hildesheim, Germany
Envelope-Open reusspa@uni-hildesheim.de (P. Reuss); klaus-dieter.althoff@dfki.de (K. Althoff)
© 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Workshop
Proceedings
http://ceur-ws.org
ISSN 1613-0073
CEUR Workshop Proceedings (CEUR-WS.org)
individualization and putting together training plans on its own is not a good option for many
people due to lack of experience and time.
In order to counteract this problem, an application was developed that suggests training plans
for users that fit their preferences and restrictions with the help of Case-based Reasoning (CBR).
In this paper, we first take a look on some existing related work (Section 2) and then describe the
basic idea and the goals of our application in Section 3. We also take a more detailed look on the
knowledge modeling of our prototype in Section 3.1 and a short overview of the implementation
in Section 3.2. The evaluation setup and the results are then discussed in Section 3.3. The paper
ends with a conclusion and an outlook on future work.
2. Related work
Using CBR to suggest training plans is not a new idea. In recent years, a number of researchers
have been working on this topic. One system was build to tailor training plans especially to
obese people. It suggest plans, that were rated as efficient to loose weight by other people with
similar characteristics. A case in this system contains information about the user like age, sex,
weight, height and the body mass index (BMI). In addition, specific measurable blood values are
stored, for example the values for cholesterol, glucose, and triglycerides. The blood values are
more important for the similarity measurement than the basic physical values, therefore they
have an higher weight.
Four types of training are available in the system: water training or swimming, concurrent
training, where aerobics and strength training are performed in parallel, electrolypolysis,
where fat zones are treated with tension and finally a combination of aerobic training and
electrolypolysis. The duration of the training and observations during the training are also part
of the solution of the case. In order to provide the experts who monitor the training sessions
with all the necessary information, the cases also contain the measured values of the persons
after the training, observations during the training and their consequences.
In the system, training plans based on user attributes are recommended in order to find
the most efficient plan for a user. The system is heavily supervised by experts and has a very
extensive data collection at the beginning, which is only possible with specialist staff. Part of
the data is determined by a blood sample. This leads to well-comparable data, which are very
specific for the purpose, in the case of weight loss, but as a consequence the system can not be
used by users on their own[4].
Another system was designed to suggest training plans for exercises in gyms either with
free weights or machines. It uses a combination of Rule-based Reasoning (RBR) and CBR and is
designed for beginners in weight training. The problem structure contains information about
the user itself as well as information about the muscles to be trained. The solution is the same
fitness recommendation as for the most similar user if there is sufficient similarity. In the event
that no sufficiently similar user is found, a second mechanism is applied, which aims to find an
exercise that fits the user. The input will then be compared with an single exercise and not a
complete training plan.
A case for an exercise has the same attributes as those that a user can specify. There is also
an attribute that indicates the level of difficulty and is matched with the activity of the user,
an attribute for sex, and an attribute that describes the necessary equipment. The necessary
equipment and health complaints are not taken into account in the calculation of the similarity.
In addition, the similarity measure uses the Sorsen-Dice coefficient[5] and therefore all attributes
are equally weighted[6].
A third system uses CBR to generate training plans for marathon runners. Based on the
previous training, the system makes predictions about the result of the marathon and can
generate plans for any desired result of the marathon. Either a training plan based on a target
time is searched or a target time based on the current training plan. Data from wearables such
as the Applewatch is used to determine a runner’s performance. For the system, individual
training activities are tracked with data at 100 meter intervals. From these activities, the values
for total distance, average speed, and both the fastest and slowest times for 10, 5 and 1km are
read.
For a case, data is aggregated per week, with the number of activities and the longest distance
being its own attributes. The speeds for the longest distance are also listed separately. For
each week, the time distance to the planned marathon is relevant, as this determines the time
period for possible training. For a given week, a case consists of the data of the current week
including the data of the longest distance regardless of the current week and the current time of
the person. For the person himself, only the sex, age and weight are used. sex is included as it
has a big influence on training performance in the marathon. Since the amount of training has
a big influence, a week always represents a case that is only compared to case from the same
week.
When a future time is determined, a runner’s data is entered in week X and the result is
his weighted average time of k most similar runners. If the target is a training plan, the input
consists of the training that the user has completed at the time and an adapted training time.
We are looking for the most similar case of a user who has completed a marathon close to the
new finish time[7].
In addition to these systems, other approaches for generating training plans can be found.
Skreik an his colleagues developed a system based on automated plan generation with a planning
domain definition language (PDDL). The system defines the current performance level of an
athlete with the goal to bring the athlete to the best level of performance for him, regarding
a given goal date. The starting level is the current state of an athlete, which is determined
by dividing the body into upper, middle and lower body and these areas then again into
biceps, triceps, deltoids, trapezius, lats, abdominal muscles, quadriceps, hamstrings, and calves.
Maximum force, endurance and power are measured and recorded, as well as general tests for
aerobic and anaerobic performance. The tests vary from exercises on a sports field to laboratory
tests.
Different tests are made for each muscle group and the end result is an average of them, thus
creating a good picture of the performance. These tests must be repeated regularly to ensure
that the values develop as planned. In order to determine the actions that are needed, users
must also indicate which aspect they would like to improve and how much they would like to
do in which time frame[8].
All these systems are good approaches to generate training plans, but are limited in several
ways, either by a specific group of users, a specific sport, or the effort for monitoring the
progress. Our application is aimed at a broader scope for users in different training states and
are training either in a gym or at home with machines, free weights, or only body weight. As a
consequence, we have a broader case structure with more attributes and a broader knowledge
base. The considered systems are a good inspiration for the required data, case structure, and
similarity measures used in our system.
3. Weight training plans with CBR
The training application should be used by people who start with weight training or already
have some knowledge and experience in weight training. And those people do not have an
expert to aid them during their training. In addition, the application should support exercises
with machines, free weights or body weight, performed in a gym or at home. Therefore, the
application has to consider information about the users, their restrictions and preferences, the
possible exercises, and the combination of exercises in training plans. We have identified six
different use(r) cases for our application to cover:
1. A beginner in sports wants to start with weight training and needs a training plan for a
whole body workout.
2. An active athlete in other sports that wants to start weight training as an addition.
3. An active athlete in weight training that wants to split his/her whole body workout into
several training plans with muscle group focus.
4. An active athlete that wants to switch from gym workout to home workout.
5. A travelling athlete does not have access to the equipment, but does not want to neglect
his workout.
6. An active athlete with progression problems wants a new different training plan.
The first and second use case deal with newcomers, while the other use cases deal with the
support of people already in training. For those who are already active, it is assumed that people
have a rough idea of fitness, but for whom a change of training plan is not trivial. For beginners,
it is about helping them to provide a suitable training plan that allows them to achieve their
goals efficiently and takes into account their possibilities. In addition, the time required to get
started can be reduced. Both points can be served by generating complete training plans.
The third use case is about facilitating changes in training. This may be because the different
muscle groups need to be trained more intensively or because there is less time available for
training and the existing plan needs to be divided into several groups. The fourth and fifth use
case deal with the support of an environment change, which can be temporary or indefinite.
Whether a new plan is necessary for the change from the gym to the home workout, or whether
only exercises need to be exchanged, depends on which plan the user already has and how
much equipment he owns.
The last use case is about users who follow a training plan but are on a plateau, and these
users can benefit from a completely new plan for the affected muscle groups, as this plan can
provide different stimuli. Even if the reasons are different, all use cases need either to start a
training plan or an adjustment that can be made depending on the extent of the changes made
by a new plan or by exchanging exercises in an existing plan.
3.1. Knowledge modeling
Initially, the system has to be filled with a lot of plans and exercises as well as potential
restrictions and preferences. The data about the exercises, plans, and restrictions of exercises
has to be collected through experts. User restrictions and preferences have to be collected by
user inputs. After the initial knowledge modeling, users can create and generate new plans for
their workout. The experts can then evaluate these training plans and decide whether or not a
new plan should be included in the knowledge base. An exercise or training plan that is not
approved can still be used by the user who created it. In Figure 1 an overview of the modeled
attributes can be seen.
Figure 1: Attributes, data types, and weight of the knowledge model
3.1.1. Knowledge model for users
Attribute age: Based on the six use cases, the knowledge model has to consider information
about potential users. The first relevant information in our context about a user is the age. The
age influences a training in several ways. In addition to the performance, age also influences
the choice of training, since the goals, especially before puberty, should be more focused on
movements rather than on strength in order to provide a solid basis for later strength training.
Especially in older age, the muscle strength of people decreases and thus the performance
and, over time, the ability to perform certain activities. For older people strength training can
lead to an improved quality of life, as it can counteract muscle loss and at the same time train
the mobility of the body [9]. Also, the effects of training plans are influenced by age, which
means that training plans should not be easily transferred to large age gaps, even if they are
not pensioners, children or young people[10].
For the similarity of age, the relationship between age and performance was considered. This
connection has been studied in the field of professional sports and different trends are emerging
for the different sports. Although the endpoints differ, the courses themselves are very similar.
In addition to athletic performance, mental performance, which can be influenced by age, is also
evaluated. Even as a result of accidents or unhealthy lifestyles, people sometimes fail to reach
their peak performance, but can improve their personal performance later in life by making
changes[11][12][13]. In summary, a 10-year age difference should be more problematic in older
age. This is because at the beginning of 20 the peak should be reached slowly and a plateau
should follow, while at 40 the extraction is already stronger and the differences per year should
be greater. The relative changes are not significantly different between the sexes, so it is not
considered a factor. This results in a similarity function for age, which calculates a similarity of
one up to a difference of two years, drops to 0.6 up to a difference of 5 years and drops to zero
at 10 years.
Attribute experience: Another relevant information is the experience of the users. In order
to recommend training plans to users that are able to perform the exercises safely and which are
at the same time appropriate in scope, the user’s experience must be determined. Without the
information at all, there is a risk of suggesting exercises that a user cannot yet perform safely,
even if they have guidance. For modeling purposes, a symbolic representation is used to map
three groups of experience: beginner, advanced, and professional. The similarity is determined
using a matrix, with beginner having some similarity to advanced athletes but no similarity to
professionals. Advanced athletes have a higher similarity to professionals than to beginners.
Attribute sex: Another relevant information for training plans in weight training is the sex
of the athletes. The sex of a person is about the physical differences in the physique of men and
women. It has been shown that the maximum strength, measured by knee extensions, that men
and women can exert is very different in adulthood and only converges with increasing age, as
the musculature of women degrades less[11]. In addition, there are differences in overall muscle
mass and especially in muscle mass in the upper body[14]. For the representation of gender,
a symbolic representation was chosen. Male, female or diverse are offered as possible values
to give as many people as possible a choice. While sports research often only differentiates
between male and female and there are relevant differences, people who feel they were born in
the wrong body are also on the spectrum between male and female. In the course of hormone
therapy, some characteristics relevant to sports would also adapt, making the value divers
meaningful[15]. For the similarity measure, it is assumed that a person who chooses divers
is someone who adjusts his gender to his self-perception through therapy and thus is of the
physical condition between the sexes. There is a simplifying assumption that there is a linear
change in the relevant characteristics, therefore diverse has a 50 percent similarity to both male
and female. This assumption will not apply to all users and people will also choose diverse who
may not be doing therapy.
Attribute weight: The body weight of a person influences the performance that can be
achieved during exercises. It can be composed of different parts of muscle mass and body fat
and has a different influence depending on this, so the weight must be put in context with the
sporting activity of a person. For body weight exercises, a high body weight can be a problem
because the muscles and joints are subjected to a higher load and therefore the number of
repetitions is lower than for people with less body weight. The similarity for the body weight
is calculated via an integer function with a linear descent of the curve until a difference of 50
kilogram is reached.
Attribute size: Under certain circumstances, body size may also be relevant for the creation
of a training plan. Not all equipment is adjustable for all heights and therefore usable. The
taller person also has a different lever for some exercises, making the repetitions for the same
weight harder for the taller person. This is especially true for body weight exercises, as a smaller
person will have less weight on the forearms for push-ups, for example, even though the two
people weigh almost the same. Due to the rather small influence of height on the design of the
training plan, a distance of 50 cm is set based on the size distribution in Germany[16], at which
height still has a similarity of 0.5. From that point on, the similarity decreases more and reaches
zero at a difference of 80 cm or more.
Attribute limitations: Two other information groups are relevant from the athletes per-
spective: the limitations or restrictions and the available equipment. An example of limitations
is the limited load-bearing capacity of joints, which means that certain exercises cannot be
performed because they cause more damage than they help. But diseases can also have an
impact on exercise selection and performance. In the context of the app, we will look at ailments
that directly relate to exercise selection in a training plan or otherwise have easily understood
influences on the acute variables of a training plan: Shoulder, upper spine, lower spine, elbows,
knees, ankles and wrists, hip joints. In addition, two medical conditions where the influence on
weight training are documented: Diabetes[17] and Heart Disease[18].
Attribute available equipment Equipment or devices represent another attribute of a user,
they describe the options they have for performing exercises. With the help of this attribute, a
training plan can be better tailored to a user, as it can take advantage of the user’s options to
achieve an efficient result. At the same time, the attribute can help the system not to suggest
training plans to the user that do not fit the user because the user does not have the necessary
equipment. The range of values includes dumbbells and barbells, pull-up bar, kettle bells, dip
bars, thera-bands, weight bench, gym, and no equipment.
In addition to the user, the training plans and exercises must also be modeled. A training plan
is composed of a set of exercises and always pursues a goal, for a desired muscle group. The
exercises in a training plan determine how long it lasts and what equipment is needed for it.
3.1.2. Knowledge model for training plans
Attribute goal: A training plan is described by three primary attributes and a list of exercises.
The primary attributes are the goal of the training plan, the muscle group(s) to be trained, the
duration to absolve all exercises of a plan, and the weight volume. A training plan needs a
goal to be effective[19]. In our context, four goals were defined for strength training: Power,
maximum strength, strength endurance and muscle mass. The similarity of the targets is
defined in a similarity matrix, where the three targets maximum strength, muscle endurance
and muscle mass still have similarities of 0.6 among themselves while power is not similar to
the other targets. The reason for this is that the first three goals can be rearranged quite well
by adaptation. For the first three goals, the same or very similar exercises can be used, with
different sets, repetitions, rests and weights, while a training plan for Power consists largely of
specific exercises.
Attribute muscle group: Another attribute that is needed is the muscle group that the
training plan should train. Since not all training plans should address all muscle groups, so that
less time needs to be invested per workout or only certain aspects of the body should be trained,
it must be clear which areas a training plan should address. The muscle groups are also shown
as symbols. For the subdivision of the muscle groups, the individual areas are considered, which
can be trained and combined with split training[9][20]. A total of nine muscle groups have
been defined: arms, chest, abdomen, legs, back, shoulders, upper body, lower body, full body. A
similarity matrix is used to calculate the similarity. Since the muscle groups cannot be adapted
by adaptation, since this would require exchanging exercises for other exercises that are not
similar to them, only matches are considered here.
Attribute duration: Each plan has a certain duration needed to complete it. The time, for a
given plan, may vary somewhat between two people. This is due to the fact that the duration of
a repetition of an exercise is not always exact or the planned breaks are not followed correctly.
The attribute is very important because the potential user of the app has a limited time budget
for which he is looking for a training plan. The duration is specified as an integer in seconds.
For the duration similarity, a maximum difference of 15 minutes is considered, above which the
similarity is zero. The similarity decreases linearly for the function until it reaches zero. The
maximum difference of 15 minutes is based on the assumption that one exercise unit lasts about
5 minutes and thus a maximum of three exercises more or less are included in the training plan.
Attribute training type: A training plan is applicable for a certain type of training. We
modeled three training types: upper body training, lower body training, and whole body
training. This way, the training plans for a specific desired training can be retrieved.
Attribute exercises: The last attribute of the training plan is a list of exercises. In order for
the application to exchange exercises, the exercises must be mapped as completely as possible
to ensure that the efficiency of a training plan suffers as little as possible.
3.1.3. Knowledge model for exercises
Attribute primary muscle group: An exercise is represented by seven attributes. One
attribute is the primary loaded muscle, which is primarily relevant for the exchange of exercises.
Like the muscle groups, these muscles are also shown as symbols, but in more detail than
the muscle groups. A total of 15 different muscles are distinguished in the attribute values:
biceps, triceps, forearms, chest, straight and lateral abs, anterior and posterior shoulders, calves,
anterior and posterior thighs, gluteus, as well as lower, middle and upper back. The similarity
measure of the attribute is modeled as for the muscle groups. For an exercise to be considered
similar, the primary muscle has to match.
Attribute secondary muscle group: In addition to a primary loaded muscle, many exercises
also have another loaded muscle that is trained as a side effect. Thereby it is again relevant for
the exchange, even if to a lesser degree. It is possible that the secondary muscle is noticeably
co-trained or that there is no good separation between primary and secondary muscle. The
values are the same as for the attribute for the primary muscle with an additional value if no
secondary muscle is used. Unlike the primary muscle, a deviation in the secondary muscle is
less bad and may even be desired to achieve a different type of exercise for the primary muscle.
Therefore, the similarity measure can be more differentiated for this attribute.
Attribute equipment: Each exercise needs an assigned piece of equipment so that the list
of equipment for a training plan can be recognized based on the exercises used. The equipment
is also needed so that when an exercise is replaced, it can be noted whether a user can perform
a particular exercise at all. For similarity, only an exact match is considered. The exception is
the equipment gym, as this attribute combines several pieces of equipment. Therefore, a gym
has a similarity of one to all items that are assumed to be in a gym.
Attribute body weight exercise: In order for the system to be appropriately extended for
body weight exercises, an attribute is defined to record whether an exercise is body weight
exercise. Since the attribute has only two values, it is represented as a Boolean.
Attribute used joints: Exercises can be divided into two categories, multi-joint and single-
joint, which describe how many joints an exercise uses. Since multi-joint exercises involve a
majority of muscles, they are usually more complicated to perform, but allow for greater weights
to be moved and thus offer an advantage for strength training. In comparison, single-joint
exercises are easier to perform and isolate a specific muscle, making them more interesting
for beginners. Since these types of exercises fulfill different roles in the context of the training
plan, it is important for swapping an exercise to maintain its characteristic. Only two types are
distinguished, so a Boolean is sufficient for representation[19][9].
Attribute muscle movement: Another relevant property for an exercise is the muscle
movement that is performed. This also has an influence on the effect that the training has on a
muscle, so this property should also be maintained. The three types of muscle movement are
not particularly important when creating a training plan, as this point has usually been taken
care of by the exercise selection. If an exercise should be replaced from a plan, it should be
replaced by an exercise that has the same movement characteristics[20]. Three movements are
distinguished: push, pull, and isometric. Due to the different nature of the exercises, which
have different muscle movements, no similarities are established between them.
Attribute explosive movement: There are exercises that involve explosive movements,
these are mostly exercises for improving power. The exercises are rarely just, faster executed
versions of the otherwise also used exercises. For power training, more specially designed
exercises are used[9]. In order to avoid exchanging a power exercise with a normal exercise for
a muscle group, this attribute is used to distinguish between these types of exercises.
3.1.4. Case structures
Based on the required attributes, two case structures are modeled: a case structure for the
training plans and one case structure for exercises. The structure for training plans contains
the attributes for the user and the training plan as a problem description and a list of exercises
as the solution. The structure for the exercises contains the described attributes as the problem
description and another exercise as the solution.
3.2. Implementation
The prototypical application was implemented as a client-server-structure with a smartphone
app as the client and the CBR system as the server. As a tool for building the CBR system, our
open access tool myCBR [21][22] was used. The application currently supports the retrieve and
reuse steps of the CBR cycle [23].
When a user creates a training plan, the CBR component attempts to determine the appropriate
intensity by searching the training plans of similar users for exercises that train the same
muscles with the same goal. After the similar users are determined, their training plans are
loaded that have the same goal as the training plan that was created. For each exercise, the
training plans are searched for all exercises that train the same primary muscle and match
the specified equipment. The exercises that do not match in the most similar training plan
can thus be replaced. If a user wants to further customize their generated training plan, they
can manually replace exercises by editing the entire training plan or they can get suggestions
by retrieving a list of individual exercises that can be integrated in the training plan. The
exercises itself are not adapted during the reuse step, but the training plans by using case-
based adaption of exercises. The current implementation of the app can be found on Github:
https://github.com/FitnessCBR/FitnessCBR.git
3.3. Evaluation
The evaluation of the prototypical system was somewhat challenging, because during the
COVID-19 pandemic the physical activity of many people was reduced and therefore finding
people performing the suggested training plans was not easy. We conducted an expert evaluation
of the generated training plans and the replacement of exercises. Two expert fitness trainers
queried training plans with the app and analyzed them. The fitness trainers has been working
for 6 respective 9 years for the company RSG Group which has been operating since 1996
and has as its core business the operation of fitness studios. The company now owns many
well-known fitness brands such as Gold’s Gym, McFit and Cyberobics. Together, more than 6
million members can be recorded in studios and digital offers. Both trainers have a licence class
B for fitness training. Each trainer queried the CBR system ten times with different user and
training plan characteristics. At first the retrieved plans were evaluated as a whole and in a
second step, specific exercises were selected for exchange and the new exercises were evaluated
in context of the training plan. The fitness experts rated the training plans and exercises on a
scale from 1 to 5, where 1 means a perfect match and 5 a totally unsuitable plan or exercise.
Table 1 shows the results of the evaluation.
The evaluation results show, that the system is capable of retrieving good training plans,
but not every time a perfect match. But a perfect match is also dependent of the context and
perspective of the user. The training plans for power training do not fit very well, but the
feedback from the trainers suggest, that the system contains to few exercises for power training.
The exchange of exercises works also good, but for a better match not only the replaced exercise
should be considered but the other exercises in the plan and the order of the exercises, too.
Because the order of an exercise in the training plan has an impact of difficulty of an exercise
that should be suggested. Overall all training plans followed established rules and contain no
surprising exercises or combinations.
Table 1
Evaluation results of the training plans and exercises by experts (E1/E2)
Training plan result(E1/E2) exchanged exercises result(E1/E2)
max strength 2/2 chest 2/3
max strength 2/2 shoulder 3/3
max strength 3/2 shoulder 3/4
power 4/4 biceps 3/3
power 3/4 legs 2/3
strength endurance 1/2 triceps 2/2
strength endurance 2/2 chest 2/2
strength endurance 2/1 legs 2/1
muscle mass 2/2 chest 2/3
muscle mass 2/3 biceps 2/2
4. Conclusion and outlook
In this paper we present an application for the generation and suggestion of training plans for
weight training in different use cases. We describe the motivation behind it and take a detailed
look on the knowledge model and the decisions behind it. We also give an short overview of
the current prototype and the first evaluation with fitness experts.
Given the current state and the results of the evaluation, we plan to extend the knowledge
model with some additional attributes like the number of sets for exercises and the order of
exercises in the training plan. The retrieval process for exercises to be exchange will also be
reworked to consider the context of the exercise in the training plan. An extended evaluation
is also undergoing with the current version of the prototype by working with a number of
non-expert athletes to evaluate the training plans. This evaluation will be repeated with the
enhanced prototype.
References
[1] W. H. Organization, Physical activity: Key facts, 2020. URL: https://www.who.int/news-room/
fact-sheets/detail/physical-activity.
[2] A. Batrakoulis, European survey of fitness trends for 2020, ACSM s Health & Fitness Journal 23
(2019) 28–35.
[3] T. Peçanha, K. F. Goessler, H. Roschel, B. Gualano, Social isolation during the covid-19 pandemic
can increase physical inactivity and the global burden of cardiovascular disease, American journal
of physiology. Heart and circulatory physiology 318 (2020) H1441—H1446.
[4] F. Lorenzi, R. Rosa, A. Peres, G. Dorneles, A. Peres, F. Ricci, Tailoring training for obese individuals
with case-based reasoning, in: Proceedings of the 8th ACM International Conference, 2017, pp.
142–147.
[5] T. J. Sorensen, A method of establishing groups of equal amplitude in plant sociology on similarity of
species content and its application to analyses of the vegetation on Danish commons, I kommission
hos E. Munksgaard, Kobenhavn, 1948.
[6] D. Salim, N. Perdana, B. Mulyawan, Application of the case based reasoning & sorensen-dice
coefficient method for fitness exercise program, IOP Conference Series: Materials Science and
Engineering 1007 (2020) 012188.
[7] C. Feely, B. Caulfield, A. Lawlor, B. Smyth, Using case-based reasoning to predict marathon
performance and recommend tailored training plans, in: Case-based Reasoning Research and
Development, 2020, pp. 67–81.
[8] T. Skerik, L. Chrpa, W. Faber, M. Vallati, Automated training plan generation for athletes, in:
Proceedings of the 2018 IEEE International Conference on Systems, Man, and Cybernetics, 2018,
pp. 3865–3870.
[9] L. Brown, N. S. . C. A. (U.S.), Strength Training, Human Kinetics, 2007. URL: https://books.google.
de/books?id=SOjE7VQ2iWkC.
[10] W. Sands, J. Wurth, J. Hewit, NSCA: Basic of strength and conditioning manual, National Strength
and Conditioning Association, 2012.
[11] R. Lindle, E. Metter, N. Lynch, J. Fleg, J. Fozard, J. Tobin, T. Roy, B. Hurley, Age and gender
comparisons of muscle strength in 654 women and men aged 20–93 yr, Journal of applied physiology
(Bethesda, Md. : 1985) 83 (1997) 1581–7.
[12] M. Huebner, D. Meltzer, A. Perperoglou, Age-associated performance decline and sex differences in
olympic weightlifting, Medicine and science in sports and exercise 51 (2019).
[13] G. Berthelot, S. Johnson, P. Noirez, J. Antero-Jacquemin, A. Marck, F.-D. Desgorces, F. Pifferi,
P. Carter, M. Spedding, A. Singh-Manoux, J.-F. Toussaint, The age-performance relationship in the
general population and strategies to delay age related decline in performance, Archives of Public
Health 77 (2019) 51.
[14] J. Harrison, Bodyweight training: A return to basics, Strength & Conditioning Journal 32 (2010)
52–55.
[15] J. Harper, E. O’Donnell, B. Khorashad, H. McDermott, G. Witcomb, How does hormone transition
in transgender women change body composition, muscle strength and haemoglobin? systematic
review with a focus on the implications for sport participation, British journal of sports medicine
55 (2021).
[16] S. R. Department, Survey about body size in germany, 2021. URL: https://de.statista.com/statistik/
daten/studie/278035/umfrage/koerpergroesse-in-deutschland/.
[17] M. Hordern, D. Dunstan, J. Prins, M. Baker, M. fiatarone singh, J. Coombes, Exercise prescription
for patients with type 2 diabetes and pre-diabetes: A position statement from exercise and sport
science australia, Journal of science and medicine in sport / Sports Medicine Australia 15 (2011)
25–31.
[18] N. Meka, S. Katragadda, B. Cherian, R. Arora, Review: Endurance exercise and resistance training
in cardiovascular disease, Therapeutic advances in cardiovascular disease 2 (2008) 115–21.
[19] S. Bird, K. Tarpenning, F. Marino, Designing resistance training programmes to enhance muscular
fitness: A review of the acute programme variables, Sports Medicine 35 (2005) 841–51.
[20] W. Kraemer, N. Ratamess, Fundamentals of resistance training: Progression and exercise prescrip-
tion, Medicine and science in sports and exercise 36 (2004) 674–88.
[21] K. Bach, K.-D. Althoff, Developing case-based reasoning applications using mycbr 3, in: Proceedings
of the 20th International Conference on Case-based Reasoning 2012, Springer, 2012, pp. 17–31.
[22] A. Hundt, P. Reuss, C. S. Sauer, T. Roth-Berhofer, Knowledge modeling and maintenance in mycbr
3, in: Proceedings of the FGWM workshop at Learning, Knowledge, Adaptation conference 2014,
Springer, 2014.
[23] A. Aamodt, E. Plaza, Case-based reasoning: Foundational issues, methodological variations and
system approaches, AI Communications 7 (1994) 39–59. URL: http://www.idi.ntnu.no/emner/it3794/
lectures/papers/Aamodt_1994_Case.pdf.