Retrieval of Similar Behaviors of Human Postural Control from the

Center of Pressure in Elderly People with Sarcopenia

Thales Vinicius de Brito Uê, Danilo Medeiros Eler

and Iracimara de Anchieta Messias

Faculty of Science and Technology, São Paulo State University, Presidente Prudente, São Paulo, Brazil

Keywords: Center of Pressure, Elderly People, Feature Extraction, Force Plate, Information Retrieval, Postural Control,

Sarcopenia, Visual Analysis.

Abstract: Human postural control acquired by a force plate is an object of study in different Healthcare areas. However,

researchers without much experience in other areas of knowledge, such as Statistics and Information

Technology, observe their data only quantitatively. Therefore, as complement to these biomechanical analyses,

this paper aims to compare, retrieve and visualize similar behaviors of the Centre of Pressure (COP) measured

according to static positions performed by elderly people with sarcopenia, before and after the application of

a muscular training intervention. For this purpose, the medial-lateral (ML) and anterior-posterior (AP)

directions of the COP's oscillations are used, respectively, as coordinates on the x- and y-axes, to which the

Fourier Transform is applied to extract features from each set of coordinates that will represent each data

collection during comparisons by the Euclidean distance metric. The acquisitions are ranked based on the

similarity they share with the one defined as query. As a result, only acquisitions of interest are retrieved.

Case studies involved comparisons of pre- and post-intervention data collections from 4 subjects performing

different static positions on the force plate. Scatter plot visualizations, combined with comparisons and

retrievals of similar behaviors among COP’s oscillations, facilitate analyses and insights regarding the

subjects' postural balance performance during force plate data collections.

1 INTRODUCTION

Force plates, as biomechanical equipment, are

responsible for measuring data and information that

reflect the human body's movement in the context of

different work and daily activities (Uê et al., 2024).

In particular, observations and evaluations of human

postural control during the performances of these

tasks are essential for providing knowledge related to

many Healthcare areas, including Physiotherapy,

Biomechanics and Ergonomics (Advanced

Mechanical Technology, Inc., 2020).

However, analyses of data acquired by force

plates are more commonly based on a quantitative

approach involving mathematical software (e.g.,

MATLAB and SPSS), which require researchers to

have prior knowledge to be able to operate them

(Dunn et al., 2017). Consequently, researchers in the

Healthcare areas with less experience in other areas

of knowledge (e.g., Statistics and Information

Technology) become dependent on the support of

https://orcid.org/0000-0002-9493-145X

others to operate the software or on a prior and basic

training (Uê et al., 2023). Hence, despite the

predominance of statistical results for biomechanical

research (e.g., assessment of static postural control),

qualitative information has its importance in

biomechanical analyses as it provides another way of

looking at a set of data (Uê et al., 2024). A visual data

analysis provides an overview of all the data collected

or the values calculated from them for the variables

of interest, displaying how they all relate, in contrast

to quantitative information that summarizes an entire

human body's performance in a single numerical

value, as is the case with calculating a statistical

average (Uê et al., 2024).

For the purpose of applying a different approach

to visually analyzing force plates data, Information

Retrieval techniques embraces the processes of

storing, organizing and representing information so

that only those of interest to the user can be easily

retrieved, providing information from documents,

web pages, multimedia objects and other forms of

Uê, T. V. B., Eler, D. M. and Messias, I. A.

Retrieval of Similar Behaviors of Human Postural Control from the Center of Pressure in Elderly People with Sarcopenia.

DOI: 10.5220/0013288500003929

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 27th International Conference on Enterprise Information Systems (ICEIS 2025) - Volume 1, pages 771-780

ISBN: 978-989-758-749-8; ISSN: 2184-4992

771

content, both structured and non-structured (Baeza-

Yates and Ribeiro-Neto, 2013). Therefore, in

principle, the resolution of problems in this area aims

to retrieve all relevant content to a user's query, while

trying to include the minimum amount of irrelevant

content. Nowadays, the scope of Information

Retrieval applications ranges from text indexing,

modeling and classification to user interfaces and data

visualization (Baeza-Yates and Ribeiro-Neto, 2013).

Additionally, it consists in an area with strong

interdisciplinary potential, extending its applications

across various domains of knowledge, including

Healthcare.

Retrieving information that is similar to the one

adopted as reference, or query, is commonly applied

in the context of data acquired over a period of time,

i.e., time series. In order to measure and rank the

similarity between sets of values of this category,

features can be extracted from them so that these

attributes represent each set during their comparisons.

Consequently, distance metrics are applied to rank the

level of similarity, based on the characteristics

obtained for each data vector.

As a method for extracting features, also referred

to as descriptors, a set of values represented in the

time domain is converted to the frequency domain.

For this purpose, the Fourier Transform provides a

means of applying these transformations, resulting in

the acquisition of the power spectral density of the

data (Quijoux et al., 2021). Once Fourier descriptors

are extracted, they compose a feature vector able to

quantitatively describe each set of values that was

initially modeled as a time series. The task of feature

extraction for later analyses and comparisons is best

performed in the frequency domain, since it allows

for greater distinctions between descriptors

representing each data vector (Villegas et al., 2024).

The aim of this paper is to compare, retrieve and

visualize similar performances of oscillations from

the Center of Pressure (COP) in elderly people with

sarcopenia during data collections on a force plate,

before and after muscular training intervention.

Therefore, the process of recognizing similar postural

balances, involving data collections from a single

subject or several ones and whether they improve or

worsen their balance after the application of the

intervention, becomes much easier and faster.

For visualization of the COP's performance

throughout the duration of a data collection on the

force plate, it is possible to combine one of its

displacement directions by the other, i.e., to

graphically display each value measured at an instant

of the acquisition in the medial-lateral (ML) direction

by its respective anterior‑posterior (AP) direction that

occurred at the same instant of time (ML x AP) (Uê

et al., 2023). Hence, the oscillations of the Center of

Pressure were represented visually through scatter

plots. This visual representation facilitates the

perception of how the data is concentrated or

scattered, indicating a good or poor postural balance,

respectively. Furthermore, scatter plots can have their

values colored according to the instant of time at

which they were measured during data collection.

For comparisons between data collections, the

COP's oscillations in their ML and AP directions

were used to be extracted features from them, after

converting each set of coordinates from the time

domain to the frequency domain. This conversion

was applied using the Fourier Transform. The

transformed x and y coordinates were then merged

into a single feature vector, or Fourier descriptors

vector, representing the acquisitions. As a result, it is

possible to compare the feature vectors that describe

each data collection to be able to identify similar

Center of Pressure’s behaviors and, therefore, retrieve

only those acquisitions that have similar postural

balances.

This paper is organized as follows: Section 2

presents research that analyze biomechanical data

acquired from force plates by using both time and

frequency domains, along with data visualization and

classification techniques; Section 3 describes the

methodology employed to achieve the aim of this

paper, including the collection of force plate data and

its steps of processing, treatment, transformation,

comparison, visualization and analysis; Section 4

provides the results obtained after applying the

developed methodology to different case studies,

regarding the postural control in elderly people with

sarcopenia at pre- and post-muscular intervention

moments, with discussions on the visual analysis of

the results. The last section, Section 5, presents the

conclusions, advantages and benefits provided by the

development and application of this paper's

methodology.

2 RELATED WORKS

Jeong and Ohno (2017) conducted an evaluation of

the differences that work experience provides in the

workers' Center of Pressure during symmetrical load

lifting with eyes closed. The first group of

participants was composed of 20 trained and

experienced subjects from a transportation company,

and the second one was composed of six university

students with no training or skills for this context. The

Wii Balance Board force plate was used to measure

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

772

the COP’s displacements, from which the velocities

of them in both time and frequency domains were

calculated. The Fourier Transform was used to

convert the velocities’ time series to the frequency

domain. For recognition of patterns and differences

between the postural balance of experienced and

non‑experienced subjects, the Linear Discriminant

Analysis (LDA) classifier was applied to the features

extracted by Fourier Transform. For implementation

of visual representations, graphs of the medial-lateral

direction of the COP's displacement by its

anterior‑posterior direction were used, along with

graphs for both time and frequency series of the

velocities reached by the COP throughout the data

collections.

Rahmati et al. (2019), based on the participation

of 40 Parkinson's patients in a balance training over

12 sessions, aimed to evaluate the effectiveness of

their training program. Authors also investigated the

neurophysiological aspects of the balance

performance in Parkinson's patients and how training

can assist in the rehabilitation of their postural

control. Data related to the Center of Pressure was

measured using a force plate, where subjects

performed two attempts under the conditions of eyes

open or closed, while standing on a rigid or foam

surface. 20 healthy subjects participated in the

experiment as a control group. To compare the results

between groups, time series for the trajectories and

velocities of the COP's displacement were visualized

and analyzed, along with the Power Spectrum

Density (PSD) of these same parameters by using the

Fourier Transform, resulting in a comparative view in

frequency domain.

Park et al. (2024) assessed the postural control of

72 subjects diagnosed with idiopathic normal

pressure hydrocephalus (iNPH) and 56 subjects who

tested positive in the cerebrospinal fluid tap test

(CSFTT). Patients were evaluated in a static posture

with their eyes open, on the day before and after the

test. A force plate was used to acquire data related to

the Center of Pressure's performance, whose

parameters were calculated in the time and frequency

domains for subsequent comparison between pre- and

post-CSFTT results. Python programming language

and its signal processing package, SciPy, were used

for the analysis. COP's trajectories in the frequency

domain were analyzed using the Fourier Transform

and Power Spectral Density. For visual analyses,

graphs were implemented for one of the COP's

displacement directions by the other, in addition to

the medial‑lateral and anterior-posterior directions

being displayed individually by both time and

frequency.

Villegas et al. (2024) employed the Wii Balance

Board force plate to measure data related to the

Center of Pressure, with the aim of identifying

patterns and distinctions between static postural

behaviors of 32 elderly people classified into three

groups: diabetics; healthy people; and those

presenting diabetes with diagnosed diabetic

neuropathy. Over 30 seconds, the subjects performed:

standing posture under the conditions of eyes open

and closed; on stable and unstable surfaces; and

performing, or not, a second cognitive task. Values

measured for COP's oscillations in the ML and AP

directions were represented by time series, from

which features were extracted to be compared and

classified using machine learning methods. The

feature extraction was also applied to COP's

parameters converted into the frequency domain,

since the spectral power is better suited to distinguish

different groups compared to temporal

characteristics. To this end, the Discrete Fourier

Transform was applied to the temporal data to

transform them into a feature vector in the spectral

power domain. Among the selected machine learning

classifiers were the K-Nearest Neighbor (K-NN)

method, which enables pattern recognition by

comparing the Euclidean distance between a data

used as a reference/query and the remaining data in

the sample.

3 MATERIALS AND METHODS

Data collections used in this paper were conducted by

Bertolini et al. (2021), using OR6-6 force plate model

from Advanced Mechanical Technology, Inc.

(AMTI). The aim of their study was to evaluate the

static postural control of elderly people with

sarcopenia, who's balance is affected by the loss of

muscle mass, before and after a 12-week muscular

training intervention (Bertolini et al., 2021). Each

subject was positioned on the surface of the force

plate and made three attempts at the static positions

of feet together (FT) and feet apart (FA) with eyes

open (EO) and eyes closed (EC) for 30 seconds; as

well as the semi‑tandem positions (ST) with eyes

open and closed; and unipodal stance with the

dominant foot and non-dominant foot, for ten seconds

(Bertolini et al., 2021). Data was collected at a

frequency of 100 Hz.

The force plate measures the ground reaction

force components and their respective torque

moments acting on the x-, y- and z-axes of an

orthogonal coordinate system (Advanced Mechanical

Technology, Inc., 2020). The components of force

Retrieval of Similar Behaviors of Human Postural Control from the Center of Pressure in Elderly People with Sarcopenia

773

and their moments are measured according to the

action performed by the subjects on the surface of the

plate. Actions can be characterized as static (e.g.,

subject standing still with changes in relation to the

positioning of their feet or to the condition of their

vision or surface) and dynamic (e.g., subject

performing a jump, walk or run). By obtaining the

force and moment values during data collections, it is

possible to calculate the behavior of the subject's

Center of Pressure over time in its medial-lateral and

anterior-posterior displacement directions,

corresponding to right-left (on the x-axis) and

front‑back (on the y-axis) displacements, respectively

(Duarte and Freitas, 2010). The COP is a measure of

positioning that has the force plate as the

biomechanical equipment commonly used to measure

it, and the observation of the COP's behavior is used

as the most significant evaluation of human static

postural control (Duarte and Freitas, 2010).

For each static position performed by the subjects

and collected before and after intervention, the values

acquired for the Center of Pressure's oscillations in its

medial-lateral and anterior-posterior directions were

used to represent, respectively, coordinates on the x-

and y-axes belonging to each point of the COP’s

displacement throughout data collection. Hence,

every static position performed by a subject on the

surface of a force plate has oscillations in the ML and

AP directions of the Center of Pressure. By

representing these displacement directions as

coordinates for the x- and y-axes, the Fourier

Transform was applied to these two sets of values to

extract descriptors from them, characterizing and

describing the COP's behavior throughout each data

collection. Therefore, it becomes possible to

distinguish or assimilate an acquisition to others that

have different or similar displacements in the x- and

y-axes. With the combination of the two sets of

Fourier descriptors, a feature vector was obtained to

represent the COP's behavior and to be used to make

similarity comparisons with feature vectors extracted

from other data collections.

As a method to compare the oscillations of the

COP obtained during the pre- and post-intervention

data collections, data was transformed into features to

describe and represent a data collection. This process

enables the calculation of a similarity distance

between acquisitions. For this purpose, the Fourier

Transform was applied to both sets of coordinates

(one with values on the x-axis and another with values

on the y-axis) and, from the resulting values, the data

extracted from each set were those that belonged to

the interval starting from the central value subtracted

by ten, and ending at that same central point added to

ten. By including the middle value, a total of 21 data

points were extracted from each set, which are

referred to as Fourier descriptors. By merging these

descriptors obtained with the Fourier Transform

applied to both sets of coordinates, a new set was

generated, named feature vector, containing 42

descriptors. The first half of values in this vector

correspond to the feature extraction from the

coordinates on the x-axis while the second half

corresponds to the same process for the y-axis.

To determine how similar the Center of Pressure's

performance is from one data collection to another,

the Euclidean distance metric was applied to compare

the feature vectors acquired from each static position.

Based on this metric, it is defined that the shorter the

distance results, the greater the similarity. And the

longer the distances, the less similar the data

collections will be to one another.

The COP in its medial-lateral and

anterior‑posterior directions of displacement was

visualized in the form of scatter plots colored to be

distinguished by subject, positioning and moment of

intervention. The displacement values in the graphs

were also individually colored to reflect the instant of

time they occurred during data collection. All steps of

data processing, treatment, analysis and visualization

were accomplished by using Python programming

language.

Finally, in addition to comparisons based on

Euclidean distance, the feature vectors acquired from

different data collections were submitted to the

PEx‑Image (Projection Explorer for Images)

software, developed by Eler et al. (2009), to visualize

how they are positioned within a feature space, where

similar vectors are represented by points close to each

other and, as the points are getting further apart, there

is a perception of how distinct the feature vectors are.

With this application, it was also possible to connect

sets of points that were close and similar with the

K‑Nearest Neighbor machine learning classifier

method. This algorithm provides connections

between the most similar neighboring acquisitions

and separates them from their more distant neighbors.

4 RESULTS ANALYSIS

As a case study to demonstrate the developed

application, the data collections acquired from a force

plate for four different elderly people with sarcopenia

were evaluated during the pre- and post-intervention

moments of a muscular training program. Each

subject performed three attempts at the static

positions of FTEO, FTEC, FAEO, FAEC, STEO and

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

774

STEC at both intervention moments. In total, 144 data

collections were gathered, from which the

displacements of the Center of Pressure in its

medial‑lateral and anterior-posterior directions were

calculated, representing the balance and postural

control behavior of the subjects during the

performance of each position. For a visual

representation of the postural behavior, scatter plots

were modeled to display the coordinates obtained for

each COP oscillation along the x-axis (medial-lateral

direction) and y-axis (anterior-posterior direction).

Figure 1: ML x AP scatter plots colored by subjects and

ranked in ascending order of Euclidean distances measured

in relation to the COP's performance during the positioning

defined as query (FTEC3 in the pre-intervention) and

compared to other acquisitions.

Figure 2: ML x AP scatter plots colored by time; with titles

colored by subjects; and ranked in ascending order of

Euclidean distances measured in relation to the COP's

performance during the positioning defined as query

(FTEC3 in the pre-intervention) and compared to other

acquisitions.

For each acquisition, the Fourier Transform was

applied to the coordinates of both x- and y-axes; the

Fourier descriptors were extracted from each set of

resulting values; the two sets were unified, defining a

feature vector to now represent the data collection;

and a visual analysis was then performed for two

cases of retrieving similar acquisitions from one

defined as the query.

In one of the cases, the aim was to retrieve data

collections containing CP displacements indicating

large oscillations, i.e., imbalances throughout time.

For this purpose, the third attempt to perform the feet

together with eyes closed (FTEC3) position from the

elderly person identified as the letter C, which was

acquired before the intervention applied to the

subjects, was used as the query. This means that its

feature vector, that represents it, was compared with

those features representing the other acquisitions by

using the Euclidean distance metric. Figure 1

illustrates the ranking in ascending order of the

distances obtained in relation to the subject's COP

behavior during the data collection used as a query,

corresponding to the first graph in the top left corner.

The remaining scatter plots represent the 8

acquisitions most similar to the query, arranged from

left to right and top to bottom in order of similarity.

Therefore, as there are larger similarity distances with

the query, these data collections will be positioned

less close to it. It is also important to emphasize the

relevance of the retrieval process, since it provides a

means for gathering acquisitions from the same

subject, even at different moments of intervention and

with different positions.

For the visualizations, as it is changed which

aspect that is being highlighted in the graphs of each

figure, it is important to emphasize that different ways

of coloring the graphs indicate that it is possible to

evidence different parts of the data collection,

facilitating the perception and analysis of the data. In

the first coloring method, visualizations were colored

red, green, blue and purple, so that each subject is

represented by one of these colors. The data

collections belonging to the elderly person, whose

one of his acquisitions was used as query, were

colored red. Figure 1 shows that the first 6 collections

most similar to the query belong to the same elderly

person (colored red). The last two acquisitions belong

to other two different subjects. As a result, more than

half of these unbalanced retrieved acquisitions belong

to the same subject.

Figure 3 shows the representations of static

positions being colored by the type of positioning

performed (second form of coloring). It can also be

noticed that the differentiation between eyes open and

closed for the same position is due to the change in

the tonality of the same color (e.g., light blue for the

feet together with eyes closed (FTEC) position and

dark blue for the feet together with eyes open (FTEO)

Retrieval of Similar Behaviors of Human Postural Control from the Center of Pressure in Elderly People with Sarcopenia

775

Figure 3: ML x AP scatter plots colored by positions and

ranked in ascending order of Euclidean distances measured

in relation to the COP's performance during the positioning

defined as query (FTEC3 in the pre-intervention) and

compared to other acquisitions.

Figure 4: ML x AP scatter plots colored by moments of

intervention and ranked in ascending order of Euclidean

distances measured in relation to the COP's performance

during the positioning defined as query (FTEC3 in the

pre‑intervention) and compared to other acquisitions.

position). Thus, nearly all retrieved acquisitions

involved feet together performances and, within this

group, there is a predominance of data collection with

eyes closed.

A third method of coloring the scatter plots

corresponds to the differentiation between pre- and

post-intervention acquisitions (Figure 4). Positions

collected before the intervention are colored blue and

those collected after intervention are in red. This

modeling illustrates a greater presence of

pre‑intervention positions in the recovered data

collections. Furthermore, it is important to notice that

the retrieval process identified many of the most

similar acquisitions to the query as being from the

same subject (Subject C), collected both pre- and

post-intervention. This consistency in retrieving the

same subject among those with most similar data

collections to the query, at any moment of

intervention, reflects the fact that the algorithm

succeeds in retrieving similarities. Another

consistency that can be observed is that, even in

different acquisitions and moments of muscular

training, subject C maintains the same behavior.

In addition, scatter plots were also modeled in

which each COP displacement value was colored

according to the time, in seconds, that it occurred

during data collection. The visualizations were

colored so that the darker colors (e.g., blue)

corresponded to the beginning of data collection,

while the lighter colors (e.g., yellow) indicated the

final instants. With this method of visualization, it is

possible to observe the temporal behavior of the COP

and at which instants (beginning, middle or end) the

displacement oscillations occur or not. Because the

graphs were already colored on a time scale, the text

titles of each acquisition were colored to distinct the

different subjects who performed it, as illustrated in

Figure 2. The texts were also colored according to the

moment of intervention and to the type of positioning

performed. By adding the time variable, the COP

oscillations can be better understood. In this case of

retrieving unbalanced data collections, the greatest

variations occur when the colors are darker (start of

the collection) and as the end of the period approaches

(lighter colors), some acquisitions manage to reduce

the amplitude of the COP displacements, i.e., the

subjects found a point of balance despite an

oscillating start.

Figure 5: Feature space, where feature vectors, representing

each acquisition, correspond to points colored by subjects

and connected based on the KNN classifier, highlighting

the query (FTEC3 in the pre‑intervention) and its most

similar neighbors.

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

776

After visualizing different styles of scatter plots,

the feature vectors for each data collection were

submitted to PEx-Image software, in which the

feature space with all acquisitions was visualized.

The application provides the multidimensional

projection technique Interactive Document Map

(IDMAP) to be used for the visualization of the

Fourier descriptors vectors within the feature space

(Minghim et al., 2006). The points representing

similar vectors in this space were connected after

applying the K-Nearest Neighbor classification

algorithm. This connection between data points,

highlighting the one representing the query and its

nearest neighbors, is illustrated in Figure 5, where

acquisitions are colored by subjects. Therefore, the

query data collection belonging to subject C is

colored green, and it has other acquisitions from the

same subject as its most similar. Additionally, there

are more green-colored points which are also closer

to the query. The largest cluster of points on the left

side of Figure 5 indicates acquisitions sharing a

greater number of similar characteristics, resulting in

a large concentration of points. This form of

visualization by displaying the feature space,

contributes to the information retrieval process, as it

provides an overview of the similarities between all

acquisitions performed by all subjects.

On the other case study, acquisitions with a higher

concentration of COP displacement values were

retrieved, indicating a more desirable balance. Figure

6 illustrates the scatter plot visualizations modeled to

represent the behavior of the Center of Pressure

measured during the executions of each static

Figure 6: ML x AP scatter plots colored by subjects and

ranked in ascending order of Euclidean distances measured

in relation to the COP's performance during the positioning

defined as query (FAEO1 in the post-intervention) and

compared to other acquisitions.

position. The positioning defined as query for the

comparisons with other acquisitions corresponded to

the first attempt at the feet apart with eyes open

(FAEO1) position performed during pre‑intervention.

Figure 7: ML x AP scatter plots colored by time; with titles

colored by subjects; and ranked in ascending order of

Euclidean distances measured in relation to the COP's

performance during the positioning defined as query

(FAEO1 in the post-intervention) and compared to other

acquisitions.

The visualizations for this case were also colored

in different ways for each figure, highlighting

different parts of the data collection to provide more

insights and an easier analysis of the data. By coloring

the visualizations to differentiate the subjects, as seen

in Figure 6, it is possible to notice a greater presence

of elderly people whose data collections are in green

and blue.

Figure 8: ML x AP scatter plots colored by positions and

ranked in ascending order of Euclidean distances measured

in relation to the COP's performance during the positioning

defined as query (FAEO1 in the post-intervention) and

compared to other acquisitions.

Retrieval of Similar Behaviors of Human Postural Control from the Center of Pressure in Elderly People with Sarcopenia

777

Figure 9: ML x AP scatter plots colored by moments of

intervention and ranked in ascending order of Euclidean

distances measured in relation to the COP's performance

during the positioning defined as query (FAEO1 in the

post‑intervention) and compared to other acquisitions.

On the other side, by coloring the graphs

according to the type of positioning, as seen in Figure

8, it is possible to observe the majority presence of

green coloring, whether in light or dark tones,

indicating the attempts to perform the feet apart

position with eyes open (FTEO) or closed (FTEC).

By coloring the scatter plots according to the

intervention moment at which each type and attempt

of positioning was performed (Figure 9), it is possible

to observe a dominant presence of post‑intervention

data collections colored in red among the first ones

retrieved, just as in the overall picture compared to

the amount of data collections in blue

(pre‑intervention). It is also possible to notice from

the retrieval results that some subjects (Subjects A

and B) maintained a consistency in their balance

behavior from pre- to post-intervention, in relation to

some of the positions, especially the FAEO one.

Those elderly people were able to achieve a more

concentrated postural balance, both before and after

muscular training.

The addition of the time variable to the modeling

of the scatter plots is illustrated by Figure 7 (with text

titles colored by subjects). The titles were also

colored by both positions and moments of

intervention. With this form of visualization for this

case study, it is possible to notice a greater balance of

the subjects due to a concentration of their COP

displacement values, occupying less area in the

graphs. Furthermore, the overlapping of the colors,

that indicate the time at which each displacement

value occurred in the collection, also indicates less

variations in the data, as the light colors (end of data

collection) mostly overlap the dark colors (beginning

of data collection).

Figure 10: Feature space, where feature vectors,

representing each acquisition, correspond to points colored

by subjects and connected based on the KNN classifier,

highlighting the query (FAEO1 in the post‑intervention)

and its most similar neighbors.

After applying the feature vectors, that were

defined for each acquisition, to PEx-Image software,

the visualization of the feature space was modeled

with the projection technique of IDMAP, as each

vector is being represented as a data point connected

to those most similar to it, through the K-Nearest

Neighbor method. Figure 10, where acquisitions are

colored by subjects, illustrates the KNN algorithm’s

results, in which only the point representing the query

vector and its most similar neighbors was selected

and highlighted. In this case, there are points with

different colors besides the green representing the

query. Therefore, there are other subjects’ data

collections very similar to subject C's acquisition

defined as query. It is also possible to observe that the

nearest neighbors to the query are in the largest

cluster of points on the left side of the image,

indicating multiple data collections from multiple

subjects that are also highly similar.

5 CONCLUSIONS AND

DISCUSSIONS

The force plate biomechanical equipment measures

data that provides information about the human body

movement, i.e., its behavior and postural control (Uê

et al., 2024). However, analyses of this data require

researchers to have experiences and a greater

understanding of other areas of knowledge in addition

to the Healthcare ones (Dunn et al., 2017). In

consequence, statistical analyses involving force

plate data are often conducted without a qualitative

view of the results to complement it (Uê et al., 2023).

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

778

A visual analysis provides an overview of the data

and how they are related to each other, without

summarizing the human body's behavior in a single

statistical value (Uê et al., 2024).

With data collected using a force plate, it is

possible to use visualizations (e.g., scatter plots) to

qualitatively observe the performance of a subject's

Centre of Pressure in its medial-lateral and

anterior‑posterior directions during a data collection

(Uê et al., 2023). From these displacement directions

acquired for each acquisition, it is possible to use

feature extraction techniques, also referred to as

descriptors, so these can be compared with other

features extracted from COP's displacements

belonging to other acquisitions (Baeza-Yates and

Ribeiro-Neto, 2013). Once distance metrics are

applied to compare data collections, these can be

ranked according to their level of similarity, allowing

the user to retrieve only acquisitions that are similar

to the one defined as their interest, which is used as

query (Baeza-Yates and Ribeiro-Neto, 2013).

The aim of this article is to compare, retrieve and

visualize, in a ranked manner, similar performances

of the COP's displacements in elderly people with

sarcopenia, during data collection on the force plate

before and after a muscular training intervention.

Consequently, the recognition of similar postural

balances is optimized and automatized for the

comparisons of acquisitions belonging to the same or

different subjects.

For comparisons between data collections, values

obtained by the oscillations of the Center of Pressure

in its ML and AP directions were used to correspond

to coordinates on the x- and y-axes, respectively.

Fourier Transform was then applied to each of these

two sets of values to extract descriptors from them,

which combined characterized and described the

behavior of the COP during an acquisition. Having

defined the sets of Fourier descriptors values (or

feature vectors) to represent each data collection,

these vectors were compared with an acquisition

defined as query. Then, the feature vectors of all

acquisitions were ranked using the Euclidean distance

metric, so similar behaviors would have smaller

distances from the query, placing them close to it. In

the opposite way, the greater the distance result, the

less proximity the data collection shares with the

query.

For the case studies, performances of four

different subjects were compared by using their three

attempts for the static positions of FTEO, FTEC,

FAEO, FAEC, STEO and STEC, in pre- and

post‑intervention moments (Bertolini et al., 2021). In

the first case, comparisons were made between data

collections that presented large oscillations during the

displacement of the COP, i.e., the values were

scattered across the area of the graph. One of these

acquisitions that involved many postural imbalances

was defined as the query, so that behaviors most close

to it were ranked and retrieved. Results from the

retrieval of data collections similar to the one defined

as query were visualized through scatter plots, where

values were colored to differentiate the subjects,

positions, and moments of intervention. Similarly, in

another visualization model, these three coloring

forms were used in the titles of each acquisition,

because the COP's oscillation values were colored

according to the instant of time in which they

occurred.

In the second case, the query was defined as an

acquisition whose COP's displacement values were

more concentrated and placed at the center of the

scatter plot, indicating a focused and desirable

balance. The same forms of coloring by subjects,

positions, moments of intervention, and time, were

also used for the visual analysis of the results for the

most well-balanced data collections retrieved based

on the one defined as query.

Based on the developed approach of analysis, it

was possible to identify that subjects A and B

presented the largest number of data collections with

a concentrated postural control and few variations in

its amplitude. Similarly, most of performances from

feet together with eyes open (FTEO) and eyes closed

(FTEC) positions were retrieved when it was desired

to observe COP's behaviors whose displacement

values were more concentrated. There was also a

more significant presence of post‑intervention

acquisitions among the first data collections ranked.

In contrast, subject C presented the greatest balance

difficulties, especially when performing the FTEC

position at both pre- and post-intervention, as his

displacement values were more dispersed across the

area of the graphs. In general, the execution of feet

together position resulted in the greatest variations in

postural control. Additionally, most of these retrieved

data collections exhibiting the subject's difficulties in

finding their balance point, were taken in the moment

prior to the intervention.

Therefore, as methods to compare performances

of the Center of Pressure acquired in each data

collection, the feature extraction and the ranking of

acquisitions according to the one defined as a base,

i.e., a query, consist of techniques that make it

possible to retrieve data collections presenting similar

performances of the subjects’ postural control. Thus,

by associating each acquisition with a visual

representation, it becomes faster and easier to

Retrieval of Similar Behaviors of Human Postural Control from the Center of Pressure in Elderly People with Sarcopenia

779

qualitatively analyze which positions, and in which

attempts the subjects performed in a more balanced

way and with fewer COP's oscillations, or also, which

positions and attempts showed the most postural

imbalances. In addition, coloring the data values

facilitates comparisons between subjects; positions;

moments before and after intervention; and time of

collection, whether for the same subject or for

comparisons of different sets of individuals who

performed actions in contact with the surface of the

force plate.

For future work, the inclusion of professionals and

researchers, who are familiar with force plates, is

essential and of extreme importance in evaluating the

proposed visual analysis method's usability. Users'

participation, based on interviews and their

perceptions on the visual results, is a fundamental

assessment metric which provides their observations,

opinions, and narratives regarding the advantages and

potential improvements of the developed approach to

analyze postural balances acquired from force plates.

ACKNOWLEDGEMENTS

This study was financed in part by the Coordenação

de Aperfeiçoamento de Pessoal de Nível Superior -

Brasil (CAPES).

REFERENCES

Advanced Mechanical Technology, Inc. (2020). AMTI

Home Page. Accessed 2 June 2020, <https://amti.biz/>.

Baeza-Yates, R., Ribeiro-Neto, B. (2013). Modern

Information Retrieval, Bookman Editora. Porto Alegre,

edition.

Bertolini, G. N., Silva, B. S. A., Santos, V. R., Messias, I.

A., Ribeiro, J. P. J., Marini, E., Gobbo, L. A. (2021).

Are bioelectrical parameters and functionality

associated with postural control in the elderly? Clinical

Biomechanics. Bristol, v. 82.

Duarte, M., Freitas, S. M. S. F. (2010). Revision of

posturography based on force plate for balance

evaluation. Rev. Bras. Fisioter. São Carlos, v. 14, n. 3.

Dunn, W., Burgun, A., Krebs, M., Rance, B. (2017).

Exploring and visualizing multidimensional data in

translational research platforms. Brief Bioinform. v. 18,

n. 6.

Eler, D. M., Nakazaki, M. Y., Paulovich, F. V., Santos, D.

P., Andery, G. F., Oliveira, M. C. F., Batista Neto, J.,

Minghim, R. (2009). Visual analysis of image

collections. The Visual Computer. v. 25, n. 10.

Jeong, H., Ohno, Y. (2017). Symmetric lifting posture

recognition of skilled experts with linear discriminant

analysis by center-of-pressure velocity. Intel Serv

Robotics. v. 10.

Minghim, R., Paulovich, F. V., Lopes, A. A. (2006).

Content-based text mapping using multi-dimensional

projections for exploration of document collections.

Proceedings of SPIE 2006. v. 6060.

Park, E., Lee, S., Jung, T.-D., Park, K.-S., Lee, J.T., Kang,

K. (2024). Changes in postural stability after

cerebrospinal fluid tap test in patients with idiopathic

normal pressure hydrocephalus. Front. Neurol. v. 15.

Quijoux, F., Nicolai, A., Chairi, I., Bargiotas, I., Ricard, D.,

Yelnik, A., Oudre, L., Bertin-Hugault, F., Vidal, P.-P.,

Vayatis, N., Buffat, S., Audiffren, J. (2021). A review

of center of pressure (COP) variables to quantify

standing balance in elderly people: Algorithms and

open-access code. Physiological Reports. v. 9.

Rahmati, Z., Schouten, A. C., Behzadipour, S., Taghizadeh,

G., Firoozbakhsh, K. (2019). Disentangling stability

and flexibility degrees in Parkinson’s disease using a

computational postural control model. Journal of

NeuroEngineering and Rehabilitation. v. 16.

Uê, T. V. B., Eler, D. M., Messias, I. A. (2023). A

Ergonomia Construtiva como Ferramenta na Interface

Participativa da Modelagem de Visualização Gráfica de

Posturas e Gestos. Proceedings of ABERGO 2023.

Hotel Majestic, Florianópolis, SC, Brazil.

Uê, T. V. B., Eler, D. M., Messias, I. A. (2024). A Visual

Analysis Approach to Static Postural Control Acquired

by a Force Plate. Proceedings of the 26th ICEIS. v. 1.

Villegas, C. M., Curinao, J. L., Aqueveque, D. C.,

Guerrero-Henríquez, J., Matamala, M. V. (2024). Data

augmentation and hierarchical classification to support

the diagnosis of neuropathies based on time series

analysis. Biomedical Signal Processing and Control. v.

95.

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

780