Time Series Prediction Models in Healthcare: Systematic Literature

Review

Zina Zammel, Nesrine Khabou, Lotﬁ Souiﬁ and Ismael Bouassida Rodriguez

ReDCAD Laboratory, ENIS, University of Sfax, Tunisia

ﬁ ﬁ

Keywords:

Time Series Prediction, Healthcare, Systematic Literature Review.

Abstract:

Technology has solved many of humanity’s complex problems. Furthermore, healthcare providers and re-

searchers are working together to achieve precision medicine, which is the goal of tailoring medical treatment

to the individual characteristics of each patient. As a result, patients will receive better care. In this context,

healthcare beneﬁts from Time Series Prediction (TSP) models to improve service levels. TSP models have

been successfully used to predict a variety of outcomes, such as patient readmission rates, disease progression,

and treatment effectiveness. This study presents a systematic literature review (SLR) focusing on TSP mod-

els in healthcare. Based on a systematic search of IEEE, Science Direct, Springer, Hyper Articles en Ligne

(HAL), and ACM, 50 articles published between 2018 and 2023 were identiﬁed. A review of predictive use

cases in healthcare and the TSP models used for them has been conducted in this paper. We classiﬁed these

models into four categories such as statistical models, Deep Learning (DL) models, Machine Learning (ML)

models and Hybrid models.

1 INTRODUCTION

Time series prediction is a broad area of research with

much potential application (Khabou et al., 2017), in-

cluding physical, environmental science and health-

care. It can be deﬁned as the process of predicting

variables and future variables using temporal data.

Time series data in healthcare have many sources.

Electronic health records (EHRs) data is a type of

Time Series (TS) data, that tracks patient health over

time. Also, electrocardiogram (ECG) data is another

type of TS used to record the electrical activity of the

heart. Wearable devices can also provide TS data.

In healthcare, TSP models involve analyzing se-

quential data, such as medical records, patient vital

signs, patient disease registries, and administrative

clinical data to predict future trends and informed de-

cisions. To the best of our knowledge, these mod-

els help in improving patient care and health man-

agement by the early decision about patient medical

situations, and the prediction of resource hospitaliza-

tion, such as the number of beds that will be needed

(Prasad et al., 2022), the medication that will be used.

In recent decades, TSP models have been extensively

improved and expanded. In statistical models, An Au-

toregressive Integrated Moving Average (ARIMA) is

a model that analyzes or predicts time series and rep-

resents the most used model in prediction. Machine

learning and deep learning represent the recent widely

utilized models for TSP. Our study aims to help read-

ers to know the various TSP models in healthcare and

the subjects that can be predicted by these models.

Subject prediction means the predictive task of TSP.

It can be any type of disease or quality of service. To

achieve these objectives, we conducted a systematic

review of the literature. This paper is organized as

follows: section 2 represents the process of our SLR.

Section 3 provides a discussion of SLR results. Sec-

tion 4 represents the learned lesson from the SLR. We

conclude with general highlights in section 5.

2 THE SLR PROCESS

The research methodology for this paper is an SLR,

which consists of three steps: (i) Deﬁnition of Re-

search Questions, (ii) Identiﬁcation of search strategy

and (iii) Selection of articles based on precise criteria.

2.1 Research Questions

The main goal of this systematic literature review is

to extract and organize the ﬁndings from research on

structured TSP models in healthcare, and to identify

1286

Zammel, Z., Khabou, N., Souiﬁ, L. and Bouassida Rodriguez, I.

Time Series Prediction Models in Healthcare: Systematic Literature Review.

DOI: 10.5220/0012465000003636

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

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 3, pages 1286-1293

ISBN: 978-989-758-680-4; ISSN: 2184-433X

Table 1: Search results by Resource.

Resource Number of papers

Springer 164

IEEE Xplore 34

ACM digital library 275

Science Direct 70

Hyper Articles en Ligne (HAL) 44

Total 587

related, future research opportunities subsequently.

The key aspects that provide the focus for research

study identiﬁcation and data extraction are speciﬁed

research question(s), as a result, we aimed to answer

the following research questions.

1. RQ1: In which case are time series prediction

models used in healthcare applications?

2. RQ2: What are the different models of Time Se-

ries Prediction in healthcare?

2.2 Search Strategy

We identiﬁed the initial studies in the database ac-

cording to the following keywords Which are split

into three groups.

• Group1: (“Time Series”)

• Group2: (”Prediction Model”,”Prediction Mod-

els‘’,‘’Forecasting Model”, ‘’Forecasting Mod-

els”)

• Group3: (“Healthcare”).

To get relevant results, the search method integrates

the essential concepts in our search query. Both sets

of keywords were combined with a Boolean search

(AND, OR) in the article search process. The ﬁ-

nal search string in this study is (“Time Series”)

AND (“Prediction model ” OR“ forecasting model”

) OR (“Prediction models” OR “forecasting models

”) AND (“ healthcare‘’)

2.3 Selection Criteria

Following the acquisition of search results from dif-

ferent Databases, the articles were selected using a

set of Inclusion and Exclusion Criteria. These criteria

were applied to help in the identiﬁcation of relevant

primary studies. Also, the purpose of these criteria

is to obtain results with more accuracy, objectivity,

and signiﬁcance for our study. The Inclusion Criteria

are: (i) The predetermined keywords exist through-

out the paper, or at a minimum in the title, keywords,

or abstract section, (ii) The paper was published in a

scientiﬁc peer-reviewed journal, (iii) Research stud-

ies that were published from January 2018 to March

2023 and (vi) Articles that are written in English lan-

guage. The Exclusion Criteria are (i) Publications that

are not related to the keywords of research questions,

(ii) Review papers, book chapters, master and Ph.D.

dissertations, (iii) Publications that are published be-

fore or on 31.12.2017 and (vi) Articles that are written

in any language other than English.

2.4 Data Extraction

For this SLR we used ﬁve research databases: HAL,

IEEE, ACM, Science Direct, and Springer. The

search was conducted between 2018 and 2023. The

execution of the deﬁned research query has selected

587 articles from the different resources summarized

in table 1. A total of 243 duplicate papers were re-

moved, leaving 344, then we applied the ﬁltering pro-

cess to ﬁnd 50 papers results for reading.

3 RESULTS AND DISCUSSION

TSP models are one of the important topics that many

studies are investigating to provide robust and re-

liable healthcare solutions and help stakeholders in

decision-making. This paper can be a good starting

point for such researchers to understand these mod-

els and review existing work related to the proposed

research questions. In this section, a discussion of

the analyzed publications was presented to answer the

proposed research questions

3.1 RQ1:In Which Case Are Time

Series Prediction Models Used in

Healthcare Applications?

Predicting is crucial in healthcare to identify possi-

ble health risks, prevent illnesses, and enhance patient

outcomes. Various healthcare prediction approaches

exist, depending on task speciﬁcation and TSP mod-

els. Related works show that many approaches are

about disease prediction and the quality of service

to improve medical care. Table 2 answers our ﬁrst

research question by listing the studies included in

the prediction task. Note that Coronavirus (COVID-

19) pandemic prediction as conﬁrmed cases, death

cases, and recoveries cases (Masum et al., 2020), is

the most studied prediction subject in recent years.

Also, epileptic seizure represents the second inten-

sive predictive task, it is a chronic brain disorder that

causes recurrent seizures (Rasheed et al., 2020). In

addition, infections can cause sepsis, the body’s ex-

treme response. Septic shock is the most severe form

Time Series Prediction Models in Healthcare: Systematic Literature Review

1287

of sepsis and one of the leading causes of death in

hospitals (Liu et al., 2014). The prediction of these

tasks involved several models that we will discuss in

the next part of our work.

3.2 RQ2:What Are the Different Time

Series Prediction Models

This section answers our second research question as

we will brieﬂy discuss TSP models used in health-

care. After reviewing the included, we found that the

asserted contributions of researchers within TSP mod-

els can be classiﬁed into statistical models, machine

learning models, Deep learning models, and hybrid

models.

3.2.1 Statistical Models

In this part, we present the statistical models namely

ARIMA and SARIMA.

• ARIMA

Statistical models commonly used for time-series

analysis include AutoRegressive Integrated Moving

Average (ARIMA). ARIMA models are a popular

method for analyzing and predicting TS data. Mod-

eling TS data with seasonality, trends, and noise is

possible with them. These models consist of three

components: autoregression, integration, and moving

average. Autoregression is used to model the cor-

relation between observation and several lag obser-

vations. TS data persistence is modeled using this

component. One of the most widely used time se-

ries models is ARIMA (Alqasemi et al., 2021). To

predict the future trajectory of COVID-19 such as ac-

tive, recovered, conﬁrmed, and death cases(Kumar

and Susan, 2020) analyzed temporal data on cumu-

lative cases from the ten countries with the highest

number of cases based on ARIMA and Prophet TSP

models. They applied statistical measures to evalu-

ate models. ARIMA performed better than Prophet

on the scale Root Mean Square Error, Root Rela-

tive Squared Error, Mean Absolute Percentage Error,

and Mean Absolute Percentage Error. However, they

claim that the correlation of other variables, such as

population density, weather, health system, and pa-

tient history, and the use of DL and artiﬁcial intelli-

gence can also improve prediction levels. In the work

of (Dash et al., 2021), ARIMA model was used for

predicting the daily-conﬁrmed cases for 90 days fu-

ture values of six worst-hit countries of the world and

six high-incidence states of India using time series

data. (Kumar et al., 2021) have proposed a TSP model

which is ARIMA for COVID-19 epidemic analysis

and predict the number of conﬁrmed cases in India

between February 2020 and April 2020. ARIMA was

applied to a wide range of time series data, includ-

ing non-stationary and irregularly spaced data. Fur-

thermore, they considered RMSE (Root Mean Square

Error) as a performance metric for prediction error.

ARIMA was suitable for time series analysis, and its

superior performance compared to other models. This

prediction model helps to assist public, private, and

government agencies in designing and implement-

ing decision-making policies. In Bangladesh (Ak-

ter et al., 2021) used a web-based electronic medi-

cal record system called ”Lifeline of Medical Data”

to compile medical data (health records and medica-

tion usage) to produce statistical graphs on medica-

tion usage and forecast outbreaks of deadly diseases

such as dengue fever. This platform is based on the

TSP model ARIMA to minimize the potential dam-

age caused by the outbreak of recurrent diseases.

• SARIMA

Seasonal ARIMA (SARIMA) is an extension of the

ARIMA model. It allows the modeling of time se-

ries with seasonal components SARIMA can help in

identifying trends and patterns in data that can be

used to make predictions about future values (Harper

and Mustafee, 2019). It is an effective tool for fore-

casting and can be used to analyze seasonal ﬂuctua-

tions. It was implemented to predict the COVID-19

outbreak conditions represented by conﬁrmed cases

and deaths in Australia, Canada, Egypt, India, the

United States, and the United Kingdom (Saad et al.,

2022). SARIMA models were identiﬁed as the

most suited for their data because of their residuals,

trends, and seasonality characteristics. (Harper and

Mustafee, 2019) have developed an application based

on the SARIMA model to predict the number of pa-

tients who would arrive at the Emergency department

shortly. The predictions from the SARIMA model

were then used to initialize the real-time simulation

model. The real-time simulation model was then used

to simulate the ﬂow of patients through the ED and to

assess the impact of different strategies for managing

patient ﬂow.

3.2.2 Machine Learning Models

Machine learning has proven to be a powerful tool

as TSP models in healthcare applications to reduce

healthcare workers’ cognitive load, by handling vast

amounts of TS data using different models. After we

review various studies, we note that Lasso Regression

(LR), Decision Tree (DT), Support Vector Machine

(SVM), and Gradient Boosting (GB) are the most

common TSP models used in the healthcare ﬁeld.

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1288

Table 2: List of Reviewed Studies Per Prediction Subject.

Subject Reference

Covid-19 (Masum et al., 2020; Kumar and Susan, 2020; Kumar et al., 2022)

Septic Schok (Hammoud et al., 2020; Hammoud et al., 2019; Kopanitsa et al., 2021)

Epileptic Seizure (Bhowmick et al., 2018; Xu et al., 2023; Cheng et al., 2021)

Diabetes (Song et al., 2019; De Falco et al., 2021)

Cervical Cancer (Yan et al., 2021)

Asthma (Do et al., 2019)

Mortality (Darabi et al., 2018)

Cardiovascular disease (Moshawrab et al., 2022; An et al., 2019; Perwej et al., 2018)

Coronary Heart disease (Li et al., 2022)

Heart failure (Balabaeva and Kovalchuk, 2019)

Alzheimer (Alberdi et al., 2018; Mukherji et al., 2022)

Parkinson’s disease (Gottapu and Dagli, 2018)

Physician burnout (Liu et al., 2022)

Strock disease (Yu et al., 2022)

Accident risk (Baek and Chung, 2021)

Lenght-of-stay (Olivato et al., 2022)

Admission / Readmission (Ali et al., 2022)

Predict pregnancy (Liu et al., 2019)

Birthrates (Alqasemi et al., 2021)

Dengue fever (Akter et al., 2021)

Heart sound recovery (Wang et al., 2020)

emergency call volume (Sanabria et al., 2021)

• LR

(Hammoud et al., 2020) have proposed a real-time

prediction system for septic shock in intensive care

units that uses patient vital and laboratory TS data in

combination with medical notes based on the Lasso

Regression algorithm. Lasso Regression helps to se-

lect the most important features for predicting septic

shock, which can improve the accuracy of the model

and reduce overﬁtting. The authors also have intro-

duced an extra hyper-parameter that allows the user

to increase one performance metric (AUC or median

detection time) at the expense of the other based on

a user-deﬁned utility, which provides more ﬂexibility

in model selection.

• DT

To optimize hospital services and resources, such

as patient beds and medical equipment available,

(Peixoto et al., 2022) have applied ﬁve ML models to

predict Intensive Care Unit patient daily admissions

to the hospital. The used models were Decision Trees,

Random Forests, and Gradients. MAE, MSE, RMSE,

and R2 were the evaluation metrics and showed that

DT performs better than the other compared models.

However, no model obtained sufﬁciently accurate re-

sults, so the exogenous variables used had a low pre-

dictive value.

• SVM

SVM performs well for TSP (Thissen et al., 2003)

(Moshawrab et al., 2022) used four artiﬁcial intelli-

gence models namely SVM, DNN, XGBoost, and a

Neural Oblivious Decision to analyze the character-

istics of heart rate variability and predict the occur-

rence of heart disease events based on a dataset of

heart rate variability features extracted from ECGs.

models were evaluated with the metrics of Accu-

racy, Precision, Recall, Speciﬁcity, Negative Predic-

tive Value NPV, and F1 Score. The SVM model

recorded accuracy, recall, and speciﬁcity results were

91.8%,96.66%, and 87.09% and it was able to pre-

dict cardiovascular disease 12 months before its onset

with higher performance than the other models.

• GBT

(Darabi et al., 2018) have applied GBT and Deep neu-

ral networks for predicting 30-day mortality risk after

admission to a single hospital’s ICUs based on EHR

data at the time of admission. They used medical em-

bedding to train the models as it substantially helped

with reducing the model dimensionality. The experi-

mental result of the two models shows that GBT has

better performance in prediction. GBT is a power-

ful model for training datasets that includes a lim-

ited number of records. GBT also can handle vari-

ous types of data, including numerical and categorical

features.

Time Series Prediction Models in Healthcare: Systematic Literature Review

1289

3.2.3 Deep Learning

DL models offer promising results for time series pre-

diction, such as automatic learning of temporal de-

pendencies and automatic handling of temporal struc-

tures such as trends and seasonality. Several different

DL models are used for TS forecasting in healthcare.

Recurrent neural networks (RNN) are well known to

achieve strong results in many studies with time se-

ries and sequential data. Long Short-Term Memory

(LSTM) and Gated Recurrent Units (GRU) are a type

of RNN that are speciﬁcally designed to handle long-

range dependencies in TS data. Convolutional Neural

Networks or CNNs, Multilayer Perceptrons (MLPs)

are types of neural networks also used in TSP.

• GRU

The study of (Yan et al., 2021) used GRUs to build a

clinical event prediction model for cervical cancer pa-

tients (Ma et al., 2018) have proposed a knowledge-

based attention model called KAME for predicting

patients’ future diagnoses based on TS data from

EHRs. KAME exploits medical knowledge in the

whole prediction process and achieves signiﬁcantly

higher prediction accuracy compared to other mod-

els using three real-world medical datasets. Addition-

ally, KAME learns interpretable representations of

medical codes and interprets the importance of each

piece of knowledge in the graph, making it more in-

terpretable and robust. However, KAME may have

limitations in terms of its generalizability to different

datasets or its ability to handle noisy or incomplete

data. Additionally, the effectiveness of KAME may

depend on the quality and completeness of the medi-

cal knowledge graph used as input.

• LSTM

This is a type of RNN that is well-suited for predict-

ing and extracting temporal features from long-time

TS data. Using multimodal time series data, (Ali

et al., 2022) have proposed a multitasking LSTM-

based DL model that predicts patient LOS as a re-

gression task and readmission within 30 days as a

binary classiﬁcation task. Bayesian optimizer was

used to optimize the model against the real dataset

of the patient’s physical activity in the hospital with

the new Bosch accelerometer sensor. The proposed

LSTM model predicts the patient’s readmission sta-

tus with a high accuracy of 94.84% and predicts the

patient’s length of stay in the hospital with a mini-

mum MSE of 0.025 and RMSE of 0.077, indicating

that it is a promising, reliable, and efﬁcient model.

Based on historical data linking asthma severity lev-

els and personalized trigger risk scores, in the work

of (Do et al., 2019) LSTM was combined with Addi-

tive Interaction Analysis of Exposures and predicted

respiratory and oxygen saturation risk scores. LSTM

allows for more accurate and personalized predictions

of asthma risk factors over time based on these risk

scores. Additionally, doctors and patients can pre-

vent asthma exacerbations by using personalized risk

scores to predict the likelihood of an attack. In the

work of (Haﬁz et al., 2020) the LSTM model used

is trained on past sales data of Analgesic medicine

in different districts of Bangladesh, and it predicts

the forecast value for the next few months. Overall,

LSTM plays a crucial role in accurately predicting the

demand for prescribed medicines in Bangladesh us-

ing an AI-based forecasting model. Based on the pre-

vious values of the signal, (Wang et al., 2020) have

proposed a method for heart sound signal recovery

that uses an LSTM prediction model based on the re-

current neuron network architecture. The complete

heart sound signal is used to implement a prediction

model to recover damaged or incomplete heart sound

signals. (Masum et al., 2020) have contributed to the

ﬁeld of epidemiology and public health by providing

insights into the spread of COVID-19 in Bangladesh

and predicting the future trajectory of the pandemic

as conﬁrmed, death and recovery cases in the country

using the LSTM model. The comparison of LSTM,

RF, and SVM assumed that LSTM achieved the best

result of prediction. So, LSTM is a perfect ﬁt for real-

time analysis.

• BiLSTM

SARSCoV2 spread was forecasted using BiLSTM

(Makarovskikh and Abotaleb, 2022). BiLSTM model

showed the power of decomposing daily SARSCoV2

data characterized by seasonality and trend. Re-

searchers have utilized a combination of wavelet en-

ergy features and BiLSTM deep learning networks to

enhance the extraction of more profound and mean-

ingful features from EEG signals to predict epileptic

seizures during sleep (Cheng et al., 2021).

• CNN

CNN is a typical network design for deep learning al-

gorithms that are utilized for tasks including image

recognition and pixel data processing (O’Shea and

Nash, 2015). CNN was able to make accurate pre-

dictions of COVID-19 for a variety of indicators, in-

cluding conﬁrmed cases, hospitalization, hospitaliza-

tion under artiﬁcial ventilation, and recoveries based

on TS data (Saad et al., 2022). To address the critical

objective of predicting the disease progression (Foo

et al., 2022) have proposed a framework called DP-

GAT. It is composed of three main modules: a re-

gion proposal module to extract ﬁne-grained regions

of interest (ROIs), a region features extraction module

to obtain features for these ROIs, as well as a graph

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1290

reasoning module for predicting disease progression.

They employed a CNN that utilized a sequence of

medical images as its input.

3.2.4 Hybrid Models

• CNN-LSTM

The approach of (Gottapu and Dagli, 2018) is the im-

plementation of a hybrid deep learning model that

combines LSTM and CNN to analyze and predict

Parkinson’s Disease. LSTM was used to identify

which features should be extracted to predict the pro-

gression of the disease. CNN was used for the seg-

mentation of the swallowtail, each voxel in the MRI

must be accurately classiﬁed to determine the swal-

lowtail’s precise form.

• DRSN-GRU

(Xu et al., 2023) have proposed combining a Deep

Residual Squeeze-and-Excitation Network (DRSN)

and a GRU to make predictions about seizures based

on a large dataset of EEG signals from epilepsy pa-

tients. Using a deep-learning neural network model,

the epilepsy signal is analyzed through TS, and

features are extracted by the hierarchical structure

formed by GRU and DRSN. Also, DRSN reduces the

difﬁculty of model training. The function of GRU is

to learn long-term dependencies in data and predict

epileptic seizures

• LSTM-MLP

The proposed approach of (Mukherji et al., 2022)

involves predicting the likelihood of an individual

developing Alzheimer’s disease based on a hybrid

model employing the LSTM model to forecast future

test results and MLP to diagnose people (Liu et al.,

2022) have implemented a framework, called HiPAL,

that uses activity logs from EHRs to learn deep repre-

sentations of physician workload and behavior. These

representations are then used to predict burnout risk.

It is based on LSTM that learns representations of

physician workload and behavior and MLP estimates

the burnout. The main beneﬁt of this framework is

that it can handle large EHR data.

• LSTM-GB

Using Knowledge Distillation based on LSTM and

GB models (Ibrahim et al., 2020) implement a reli-

able system namely KD-OP to predict adversity in-

dicated by death, ICU admission, and readmission.

It is an ensemble of the dynamic learner and the

static learner. The ensemble is trained using a tech-

nique of knowledge distillation, which allows the dy-

namic learner to transfer their knowledge to the static

learner. The dynamic learner uses LSTM to learn

from a patient’s physiology time series. LSTM net-

work can capture the temporal patterns in the data,

which can help predict adverse outcomes. The static

learner is a gradient-boosting model that uses static

features to estimate the risk of adversity. The gradient

boosting model can combine the static features in a

way that is not possible with a single feature.

• EMD-LSTM

(Song et al., 2019) have proposed a hybrid EMD-

LSTM model to predict patient blood glucose for 30

- 120 mins based on ECG. EMD decomposes TS of

glucose measurements into empirical modes and a

residual sequence, a different intrinsic mode function

IMF and remove the noise by reasonably choosing

IMF to Reconstruct the signal. LSTM trained each

IMF to predict the patient blood glucose level. This

model has higher accuracy than the LSTM model and

can cope with the rapid change in trends in blood glu-

cose levels.

4 LEARNED LESSONS

SLR process facilitates the research of the important

and relevant studies that we need to answer our re-

search questions. In addition, there are various types

of TSP models used to predict healthcare outcomes.

These include the length of stay, admission and read-

mission rates in hospitals, Parkinson’s disease, car-

diovascular disease, mortality rate, and risk of adverse

events. Furthermore, TSP models improved health-

care services by early warning of many events and

problems. Also, machine learning and deep learn-

ing TSP models are the most used in the healthcare

ﬁeld. Last but not least, the most commonly used

Deep learning TSP model in healthcare is LSTM.In

statistical models, ARIMA is prevalent in extant lit-

erature. Using CNN, diseases can be diagnosed and

detected effectively, and the prediction results can be

improved by combining them with RNN models.

5 CONCLUSION

This paper provides an overview of time series pre-

diction models and the predictive cases in healthcare.

For profound explanation, we categorize these mod-

els according to their types into statistical, machine

learning, deep learning, or hybrid models. By answer-

ing the ﬁrst research question, it has been shown that

Comorbidities are the most considered research topic.

Consequently, time series prediction models are cru-

cial for monitoring the health of patients.

Time Series Prediction Models in Healthcare: Systematic Literature Review

1291

ACKNOWLEDGEMENTS

This work was partially supported by the LABEX-TA

project MeFoGL: ”M

ethodes Formelles pour le G

enie

Logiciel”

REFERENCES

Akter, K., Islam, M., and Kabir, K. H. (2021). Analysis

of electronic medical records to forecast probable dis-

ease outbreaks in bangladesh. 2021 IEEE Region 10

Symposium (TENSYMP), pages 1–5.

Alberdi, A., Weakley, A., Schmitter-Edgecombe, M., Cook,

D. J., Aztiria, A., Basarab, A., and Barrenechea, M.

(2018). Smart home-based prediction of multidomain

symptoms related to alzheimer’s disease. IEEE jour-

nal of biomedical and health informatics, pages 1720–

1731.

Ali, S., El-Sappagh, S., Ali, F., Imran, M., and Abuhmed,

T. (2022). Multitask deep learning for cost-effective

prediction of patient’s length of stay and readmission

state using multimodal physical activity sensory data.

IEEE Journal of Biomedical and Health Informatics,

pages 5793–5804.

Alqasemi, F., Salah, A.-H., Alsurori, M., Hadwan, M., Al-

jaberi, Z., and Aqlan, A. (2021). Birthrates forecast-

ing based on artiﬁcial neural networks versus time se-

ries models. In 2021 1st International Conference on

Emerging Smart Technologies and Applications (eS-

marTA), pages 1–4. IEEE.

An, Y., Huang, N., Chen, X., Wu, F., and Wang, J. (2019).

High-risk prediction of cardiovascular diseases via

attention-based deep neural networks. IEEE/ACM

transactions on computational biology and bioinfor-

matics, pages 1093–1105.

Baek, J.-W. and Chung, K. (2021). Accident risk predic-

tion model based on attention-mechanism lstm using

modality convergence in multimodal. Personal and

Ubiquitous Computing, pages 1–11.

Balabaeva, K. and Kovalchuk, S. (2019). Comparison

of temporal and non-temporal features effect on ma-

chine learning models quality and interpretability for

chronic heart failure patients. Procedia Computer Sci-

ence, pages 87–96.

Bhowmick, A., Abdou, T., and Bener, A. (2018). Predic-

tive analytics in healthcare epileptic seizure recogni-

tion. In Proceedings of the 28th Annual International

Conference on Computer Science and Software Engi-

neering, pages 323–330.

Cheng, C., You, B., Liu, Y., and Dai, Y. (2021). Patient-

speciﬁc method of sleep electroencephalography us-

ing wavelet packet transform and bi-lstm for epileptic

seizure prediction. Biomedical Signal Processing and

Control, page 102963.

Darabi, H. R., Tsinis, D., Zecchini, K., Whitcomb, W. F.,

and Liss, A. (2018). Forecasting mortality risk for

patients admitted to intensive care units using machine

learning. Procedia Computer Science, pages 306–313.

Dash, S., Chakraborty, C., Giri, S. K., Pani, S. K., and

Frnda, J. (2021). Bifm: Big-data driven intelligent

forecasting model for covid-19. IEEE Access, pages

97505–97517.

De Falco, I., Cioppa, A. D., Marcelli, A., Scafuri, U., Stel-

laccio, L., and Tarantino, E. (2021). Prediction of

personalized blood glucose levels in type 1 diabetic

patients using a neuroevolution approach. In Pro-

ceedings of the Genetic and Evolutionary Computa-

tion Conference Companion, pages 1708–1716.

Do, Q., Doig, A., Son, T. C., and Chaudri, J. (2019). Pre-

dicting lung healthiness risk scores to identify prob-

ability of an asthma attack. Procedia Computer Sci-

ence, pages 424–431.

Foo, A., Hsu, W., Lee, M. L., and Tan, G. S. W. (2022).

Dp-gat: A framework for image-based disease pro-

gression prediction. Proceedings of the 28th ACM

SIGKDD Conference on Knowledge Discovery and

Data Mining.

Gottapu, R. D. and Dagli, C. H. (2018). Analysis of parkin-

son’s disease data. Procedia Computer Science, pages

334–341.

Haﬁz, M., Sazzad, M. S. I., Hasan, K. I., Hasnat, J., and

Mishu, M. C. (2020). Predicting the demand of pre-

scribed medicines in bangladesh using artiﬁcial in-

telligent (ai) based long short-term memory (lstm)

model. Proceedings of the International Conference

on Computing Advancements.

Hammoud, I., Ramakrishnan, I. V., and Henry, M. (2019).

Early prediction of sepsis using gradient boosting de-

cision trees with optimal sample weighting. 2019

Computing in Cardiology (CinC), pages 1– 4.

Hammoud, I., Ramakrishnan, I. V., Henry, M., and Morley,

E. (2020). Multimodal early septic shock prediction

model using lasso regression with decaying response.

2020 IEEE International Conference on Healthcare

Informatics (ICHI), pages 1–3.

Harper, A. and Mustafee, N. (2019). A hybrid modelling

approach using forecasting and real-time simulation to

prevent emergency department overcrowding. In 2019

Winter Simulation Conference (WSC), pages 1208–

1219.

Ibrahim, Z. M., Bean, D. M., Searle, T., Qian, L., Wu,

H., Shek, A., Kraljevic, Z., Galloway, J. B., Norton,

S., Teo, J. T. H., and Dobson, R. J. B. (2020). A

knowledge distillation ensemble framework for pre-

dicting short- and long-term hospitalization outcomes

from electronic health records data. IEEE Journal of

Biomedical and Health Informatics, pages 423–435.

Khabou, N., Bouassida Rodriguez, I., and Jmaiel, M.

(2017). A novel analysis approach for the design and

the development of context-aware applications. Jour-

nal of Systems and Software, pages 113–125.

Kopanitsa, G. D., Metsker, O. G., Paskoshev, D., and

Greschischeva, S. (2021). Identiﬁcation of risk factors

and prediction of sepsis in pregnancy using machine

learning methods. Procedia Computer Science.

Kumar, N. and Susan, S. (2020). Covid-19 pandemic pre-

diction using time series forecasting models. 2020

11th International Conference on Computing, Com-

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

1292

munication and Networking Technologies (ICCCNT),

pages 1–7.

Kumar, R., Jain, A., Tripathi, A., and Tyagi, S. (2021).

Covid-19 outbreak: An epidemic analysis using time

series prediction model. 2021 11th International Con-

ference on Cloud Computing, Data Science & Engi-

neering (Conﬂuence), pages 1090–1094.

Kumar, Y., Koul, A., Kaur, S., and Hu, Y.-C. (2022). Ma-

chine learning and deep learning based time series

prediction and forecasting of ten nations’ covid-19

pandemic. SN Computer Science, page 91.

Li, W., Zuo, M., Zhao, H., Xu, Q., and Chen, D. (2022).

Prediction of coronary heart disease based on com-

bined reinforcement multitask progressive time-series

networks. Methods, pages 96–106.

Liu, B., Shi, S., Wu, Y., Thomas, D., Symul, L., Pierson, E.,

and Leskovec, J. (2019). Predicting pregnancy using

large-scale data from a women’s health tracking mo-

bile application. In The World Wide Web Conference,

pages 2999–3005.

Liu, H., Lou, S. S., Warner, B. C., Harford, D. R., Kan-

nampallil, T., and Lu, C. (2022). Hipal: A deep

framework for physician burnout prediction using ac-

tivity logs in electronic health records. arXiv preprint

arXiv:2205.11680.

Liu, V. X., Escobar, G. J., Greene, J. D., Soule, J., Whippy,

A., Angus, D. C., and Iwashyna, T. J. (2014). Hospi-

tal deaths in patients with sepsis from 2 independent

cohorts. JAMA, pages 90–2.

Ma, F., You, Q., Xiao, H., Chitta, R., Zhou, J., and Gao,

J. (2018). Kame: Knowledge-based attention model

for diagnosis prediction in healthcare. Proceedings of

the 27th ACM International Conference on Informa-

tion and Knowledge Management.

Makarovskikh, T. A. and Abotaleb, M. S. A. (2022). Hyper-

parameter tuning for long short-term memory (lstm)

algorithm to forecast a disease spreading. 2022 VIII

International Conference on Information Technology

and Nanotechnology (ITNT), pages 1–6.

Masum, A. K. M., Khushbu, S. A., Keya, M., Abujar, S.,

and Hossain, S. A. (2020). Covid-19 in bangladesh:

a deeper outlook into the forecast with prediction of

upcoming per day cases using time series. Procedia

Computer Science, pages 291–300.

Moshawrab, M., Adda, M., Bouzouane, A., Ibrahim, H.,

and Raad, A. (2022). Cardiovascular events predic-

tion using artiﬁcial intelligence models and heart rate

variability. Procedia Computer Science, pages 231–

238.

Mukherji, D., Mukherji, M., and Mukherji, N. (2022). Early

detection of alzheimer’s disease using neuropsycho-

logical tests: a predict–diagnose approach using neu-

ral networks. Brain Informatics, pages 1–26.

Olivato, M., Rossetti, N., Gerevini, A. E., Chiari, M.,

Putelli, L., and Serina, I. (2022). Machine learn-

ing models for predicting short-long length of stay of

covid-19 patients. Procedia Computer Science, pages

1232–1241.

O’Shea, K. and Nash, R. (2015). An introduction to convo-

lutional neural networks. ArXiv.

Peixoto, D., Barbosa, A., Peixoto, H., Lopes, J., Guimar

aes,

T., and Santos, M. (2022). Predictive analytics for

hospital inpatient ﬂow determination. Procedia Com-

puter Science.

Perwej, Y., Parwej, F., and Akhtar, N. (2018). An intelligent

cardiac ailment prediction using efﬁcient rock algo-

rithm and k-means & c4. 5 algorithm. European Jour-

nal of Engineering and Technology Research, pages

126–134.

Prasad, V. K., Bhattacharya, P., Bhavsar, M. D., Verma, A.,

Tanwar, S., Sharma, G., Bokoro, P. N., and Sharma, R.

(2022). Abv-covid: An ensemble forecasting model to

predict availability of beds and ventilators for covid-

19 like pandemics. IEEE Access, pages 74131–74151.

Rasheed, K., Qayyum, A., Qadir, J., Sivathamboo, S.,

Kwan, P., Kuhlmann, L., O’Brien, T. J., and Razi,

A. (2020). Machine learning for predicting epileptic

seizures using eeg signals: A review. IEEE Reviews

in Biomedical Engineering, pages 139–155.

Saad, N. G. E.-D., Ghoniemy, S. S., Faheem, H. M.,

and Seada, N. A. (2022). An evaluation of time

series-based modeling and forecasting of infectious

diseases progression using statistical versus compart-

mental methods. 2022 5th International Conference

on Computing and Informatics (ICCI), pages 263–

273.

Sanabria, E., Lam, H., de Larrea, E. L., Sethuraman, J., Mo-

hammadi, S., Olivier, A., Smyth, A. W., Dolan, E. M.,

Johnson, N. E., Kepler, T. R., et al. (2021). Short-

term adaptive emergency call volume prediction. In

2021 Winter Simulation Conference (WSC), pages 1–

12. IEEE.

Song, W., Cai, W., Li, J., Jiang, F., and He, S. (2019). Pre-

dicting blood glucose levels with emd and lstm based

cgm data. In 2019 6th International Conference on

Systems and Informatics (ICSAI), pages 1443–1448.

IEEE.

Thissen, U., Van Brakel, R., De Weijer, A., Melssen, W.,

and Buydens, L. (2003). Using support vector ma-

chines for time series prediction. Chemometrics and

intelligent laboratory systems, pages 35–49.

Wang, Z.-H., Horng, G.-J., Hsu, T.-H., Aripriharta, A., and

Jong, G.-J. (2020). Heart sound signal recovery based

on time series signal prediction using a recurrent neu-

ral network in the long short-term memory model. The

Journal of Supercomputing, pages 8373–8390.

Xu, X., Zhang, Y., Zhang, R., and Xu, T. (2023). Patient-

speciﬁc method for predicting epileptic seizures based

on drsn-gru. Biomedical Signal Processing and Con-

trol, page 104449.

Yan, Y., Zhao, K., Cao, J., and Ma, H. (2021). Prediction

research of cervical cancer clinical events based on re-

current neural network. Procedia Computer Science,

pages 221–229.

Yu, J., Park, S., Ho, C. M. B., Kwon, S.-H., Cho, K.-H., and

Lee, Y. S. (2022). Ai-based stroke prediction system

using body motion biosignals during walking. The

Journal of Supercomputing, pages 1–23.

Time Series Prediction Models in Healthcare: Systematic Literature Review

1293