Hospital Readmission Risk Prediction Using Ensemble Learning

Mangalgouri P Kademani

, Yuvaraj P Rathod

, Shruti Nagave

, Omkar Harlapur

Uday Kulkarni

and Shashank Hegde

School of Computer Science & Engg, KLE Technological University, Hubballi, India

Keywords:

Ensemble Learning, Multilayer Perceptron, XG-Boost, Catboost, Healthcare.

Abstract:

The study focuses on features that affect of hospital readmission’s and explores how advanced machine learn-

ing algorithms can predict the chances of hospital readmission’s. Readmissions are caused by early patient

discharge, improper discharge planning, and lack of treatment, which lead to de-creased health outcomes, and

higher costs. In this study, the patient data is used from the CMS Hospital Readmissions Reduction Program

to create prediction models for hospital readmission risk. which includes over 18774 records and 12 columns

from 2019 to 2022. The machine learning models, such as MLP, XGBoost, CatBoost, and ensemble, were used

to improve the prediction’s. Where MLP achieved the accuracy of 82.69%, and XGBoost and CatBoost out-

performed MLP with scores of 85.43% and 86.50%. The accuracy of 87.08% is achieved by ensemble model,

which combined the output of all base model’s prediction outputs. Performance matrices which includes

precision, recall, F1-score were evaluated in addition to accuracy, the ensemble model obtained precision of

87.48%, recall of 87.08% , and F1-score of 86.38%. The outcomes show the result of the ensemble approach

in resolving the complex issue of hospital readmission prediction.

1 INTRODUCTION

Our topic of discussion focuses on the prediction of

hospital readmissions, a critical task in healthcare

care aimed at improving patient outcomes and re-

ducing costs. The complexity of health care data,

including missing values, discrepancies, and the in-

teraction of several readmission-causing factors(Zhou

et al., 2023), makes it difﬁcult to effectively estimate

patient readmission risk despite continuous attempts

to reduce readmission. We can improve the accu-

racy and robustness of the prediction by using ma-

chine learning techniques(Rizinde et al., 2024). In

this various machine learning models and methods

are investigated that might handle a range of health-

care datasets. Gradient boosting and deep learning are

two types of machine learning models that are popu-

lar because of their exceptional results. In order to

predict hospital readmission’s, researchers have also

looked into deep learning(Lopez et al., 2023) and

https://orcid.org/0009-0000-3214-5042

https://orcid.org/0009-0004-2635-9975

https://orcid.org/0009-0005-5928-8242

https://orcid.org/0009-0006-6302-7093

https://orcid.org/0000-0003-0109-9957

Gradient boosting techniques (Slezak et al., 2021;

Kalusivalingam et al., 2012). Such as ensemble learn-

ing approach, which combines the Multilayer Percep-

tron (MLP) (Teo et al., 2023; Ti’jay Goudjerkan, ),

XGBoost (Chen et al., 2023; Hidayaturrohman and

Hanada, 2024), and CatBoost (Safaei et al., 2022;

Quan and Gopukumar, 2023) models, which can do

the better readmission prediction.

Ensemble Learning (Mahajan and Ghani, 2019;

Turgeman and May, 2016) is an effective machine

learning technique that strengthens accuracy and con-

sistency by combining the predictions of several mod-

els. This approach involves training several models

and aggregating their outputs to address a real-world

such as predicting hospital readmission’s, where ac-

curate risk analysis is essential for patient care and re-

source allocation, this process involves training multi-

ple models and combining their outputs, to efﬁciently

process and analyze patient data. MLP takes care of

the non-linear relationships. XGBoost and CatBoost

are also great models for structured data especially

because they can handle categorical features distantly

better than other models. The Figure 1 explains a

pipeline that begins with data collection, moves on

to pre-processing and feature extraction, and ends

with encoding for machine learning-based readmis-

820

Kademani, M. P., Rathod, Y. P., Nagave, S., Harlapur, O., Kulkarni, U. and Hegde, S.

Hospital Readmission Risk Prediction Using Ensemble Learning.

DOI: 10.5220/0013603200004664

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

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 2, pages 820-826

ISBN: 978-989-758-763-4

Figure 1: Pipeline of proposed methodology

sion prediction in hospitals. The set is divided into

train and test. Various models, such as MLP, XG-

Boost, and Catboost, are trained. The accuracy of

these models is then veriﬁed using their respective

results on the test set. Finally, the system uses an

ensemble learning technique, which combines the re-

sults of multiple models to improve the prediction of

hospital readmission.

The paper is divided into 5 sections listed below:

With an overview of Several methods for group learn-

ing, such as the functions MLP, XGBoost, and Cat-

Boost, Section 2 describes the algorithms for machine

learning that are currently available for hospital read-

mission prediction. The process of preparing patient

data, training models, and combining their predictions

using ensemble methods like voting or weighted av-

eraging to produce the ﬁnal result is covered in Sec-

tion 3. The experimental results are presented in Sec-

tion 4, which compares a performance of ensemble

model with individual models on important metircs

such as F1-score, recall, and accuracy. Section 5 gives

additional details regarding the results implications

and future approaches for developing strong ensem-

ble learning techniques to improve hospital readmis-

sion prediction are also included in this.

2 BACKGROUND STUDY

Predicting hospital readmissions is a crucial ﬁeld

of healthcare analysis that has been deeply researched

through different methods of machine learning. Be-

cause to their basic analysis and implementation, tra-

ditional models such as logistic regression(Leonard

et al., 2022) have been used frequently. When there is

a clear correlation between the input factors (such as

age, clinical history, etc.) and the output (readmission

risk), the linear model known as logistic regression

performs well. However, traditional models may ﬁnd

it difﬁcult to represent the complex and non-linear in-

teractions between variables seen in healthcare data.

For instance, non-linear relationships that are difﬁcult

for linear models to accurately represent may develop

from interactions between different medical disorders

and treatments. As a result, these models frequently

lack predictive ability when dealing with the complex

of healthcare datasets.

In the area of hospital readmission prediction, ef-

fective tree-based algorithms like XGBoost (Hiday-

aturrohman and Hanada, 2024; Chen et al., 2023) and

CatBoost (Safaei et al., 2022; Quan and Gopukumar,

2023) have come up. To efﬁciently manage structured

data with missing values and complex feature interac-

tions, XGBoost applies gradient boosting. With its or-

dered boosting technique, CatBoost improves at cate-

gorical features without the need for any preprocess-

ing. While these models have shown promise, their

complexity in computation is frequently a challenge

in situations with limited resources or in applications

in real time.

Deep learning approaches, such as Recurrent Neu-

ral Network(RNN) (Chopra et al., 2017) and MLP

(Ti’jay Goudjerkan, ; Teo et al., 2023) , offer accu-

rate techniques for handling big and complex datasets.

The patient data’s cyclic patterns and non-linear re-

lationships can be captured by such models. How-

ever, many factors preventing their broader clinical

use include high computational costs, signiﬁcant pre-

processing needs, and limited comprehension.

The strengths of many models have demonstrated

that ensemble learning techniques (Mienye and Sun,

2022; Yu and Xie, 2019) can signiﬁcantly increase

predictive performance. In order to increase stabil-

ity and decrease variation, techniques such as vot-

ing, stacking, and bagging combine predictions from

various models. However, studies have shown that

ensembles frequently perform better than individual

models when managing the complexity of healthcare

datasets. A number of current ensemble approaches

may not be efﬁcient, because they do not have enough

variance among base models.

To predict hospital readmissions, other machine

learning techniques such as Naive Bayes (Rao and

Battula, 2019), Random Forests (Bleich et al., 2021;

Kalusivalingam et al., 2012), and Support Vector Ma-

chine(SVM) (Wang and Paschalidis, 2019) have also

been used. These methods work well for use with

smaller datasets or for speciﬁc applications, however

they are unlikely to deal with the huge quantities of

complex healthcare data.

Although the previously discussed research are

helpful, there are circumstances where they fall short

in terms of generalization, data handling, and model

interpretability. It is challenging to deal with limited,

imbalanced datasets and categorical features. Neural

networks and other high-accuracy models frequently

Hospital Readmission Risk Prediction Using Ensemble Learning

821

lack transparency when making medical decisions.

This study uses a new ensemble approach com-

bines CatBoost, XGBoost, and MLP to overcome

the issues. The strengths of each model are as fol-

lows: The ﬁrst model is CatBoost, it performs well at

handling categorical data, this method of use several

weak learners to make predictions is represented by

the CatBoost architecture in Figure 2. The ﬁrst step

involves processing the input training data and assign-

ing weights (W1, W2, to Wn). After that, each weak

learner (L1, L2, to Ln) uses the scattered features to

produce a prediction. All of the predictions outcomes

are saved and integrated to create the ﬁnal output pre-

diction.

Figure 2: CatBoost Architecture Diagram.

The Second Model used is XGBoost, in which it

structures data and feature relations. In XGBoost the

dataset is processed by dividing it into several subsets

(D1, D2, to Dn) using the provided XGBoost archi-

tecture in Figure 3. Individual decision trees (shown

by the circles) are then applied to each subset, pro-

ducing results that belong to those subsets (Result 1,

Result 2, to Result n). The combined ﬁnal predic-

tion is produced by adding the individual outcomes

from each tree.

And the third model is MLP, it captures nonlin-

ear relationships. The Figure 4 shows the architecture

of MLP which includes several stages, comprising an

input layer, an output layer, and one or more layers

that are hidden. Every layer is completely linked to

every other layer, and the weight of each link varies

throughout training. In the ensemble learning (Ma-

hajan and Ghani, 2019; Turgeman and May, 2016),

weighted averaging is used to improve comprehen-

sion, accuracy, and strength. This work provides an

Figure 3: XGBoost Architecture Diagram.

Figure 4: MLP Architecture Diagram.

efﬁcient and scalable solution to various healthcare

scenarios by improving the ability to generalize and

clinical use of hospital readmission prediction mod-

els through testing this approach on a difﬁcult, large

dataset.

3 PROPOSED METHODOLOGY

The proposed methodology includes combining

each model’s predictions to create the ensemble’s

ﬁnal output. Accuracy, precision, recall, and F1-score

are performance metrics that are used to evaluate the

ensemble model towards individual models.

3.1 Methods and Techniques

The proposed research improve the predicted accu-

racy and reliability of various machine learning mod-

els by combining their abilities. These are the models

that were utilized are: MLP, XGBoost, CatBoost. In

order to combine the predictions of all three models,

the ensemble model uses weighted averaging to com-

bine their results.

MLP: The feedforward neural network known as

the Multilayer Perceptron is ideal for capturing non-

linear interactions.

INCOFT 2025 - International Conference on Futuristic Technology

822

Input layer: Takes Scaled feature vector.

Hidden layer: The equation 1 uses the ReLU activa-

tion function to represent the output h

of the j-th neu-

ron in a neural network. In addition to adding a bias

, it calculates the weighted sum of inputs x

with ap-

propriate weights w

i j

. The output is set to 0 if the sum

is negative, and passes the sum unchanged otherwise.

= max

∑

i=1

i j

+ b

(1)

Output layer: The equation 2 represents the sig-

moid function. In this case, z represents the linear

combination of input features, it is usually written as

z = w

X + b, where w are weights and b is the bias.

The output, compressed by the sigmoid function, lies

between 0 and 1, representing the probability of the

positive group (y = 1).

P(y = 1 | X) =

1 + e

−z

(2)

XGBoost: It is a very powerful gradient boosting al-

gorithm that works well with feature-level interaction

and structured data. The objective function is given

by equation 3. The ﬁrst term

∑

i=1

ℓ(y

, ˆy

),is the loss

function that calculates the difference between the

predictions ( ˆy

) and true labels (y

). The second term

∑

m=1

Ω(T

), add a regularization feature that reduces

the complexity of M trees (T

), supporting simpler

models and minimizing overﬁtting.

L =

∑

i=1

ℓ(y

, ˆy

) +

∑

m=1

Ω(T

) (3)

Tree Weight Update: The weights are adjusted by

equation 4. The numerator

∑

collects the gradi-

ents, and the denominator

∑

indicates the curvature,

providing stability during optimization. This formula

minimizes the total loss by modifying the leaf’s con-

tribution to the prediction in the best possible way.

= −

∑

i=1

+ λ

∑

i=1

(4)

CatBoost: It is Not really designed for a lot of pre-

processing previously, but optimized for categorical

features. Where equation 5 shows the weighted to-

tal of the outputs from multiple base learners T

(x),

where α

are each model’s weights, as well as the ﬁ-

nal prediction f (x). Here, M is the total number of

base models, and the prediction from the m-th model

is shown by T

(x).

f (x) =

∑

m=1

(x) (5)

Ensemble Learning: To leverage the individual mod-

els, predictions are combined using weighted averag-

ing by equation 6. Where the weights given to each

model are represented by w

. The predictions of each

speciﬁc model, Model

(x), are combined to create the

result of the ensemble.

EnsemblePrediction =

∑

i=1

· Model

(x) (6)

Model Evaluation: In this, performance of models are

evalued using a variety of metrics, including accuracy,

precision, recall, and F1-score. The performance of

an ensemble model is compared to that of an individ-

ual model in order to identify improvements through

the combination of multiple models.

The Ensemble Learning Workﬂow for hospital pa-

tient readmission prediction is described in Algorithm

1. To improve prediction accuracy, CatBoost, XG-

Boost, and MLP are used in the suggested ensem-

ble learning technique. First, numerical features are

scaled for consistency, missing values are handled,

and categorical features are encoded. After that, the

dataset is used to train each model separately. Predic-

tions are produced for the test data following training.

Higher weights are given to models that perform bet-

ter after each model’s performance is assessed using

the F1-score. Individual model outputs are weighted

and added together to determine the ﬁnal ensemble

prediction. This method ensures a predictive model

that is more reliable and accurate.

4 RESULTS

The results displays performance evaluation for

different machine learning models created using hos-

pital readmission risk. The performance comparison

of an ensemble model, XGBoost, CatBoost, and MLP

is shown using accuracy, precision, recall, and F1-

score. AUC values and ROC curves are also used to

assess predictions. The experiments were carried out

on Google Colab, for effective computing and reliable

model training for high predictive performance.

4.1 Dataset Description

The Hospital Readmission dataset which is used

for this study is based on CMS Hospital Readmission

Reduction Program (Kahn III et al., 2023) which in-

cludes over 18774 records and 12 columns from 2019

to 2022. And among other metrics, the data taken

along the predicted readmission rate, expected read-

mission rate, and excess readmission ratio’s values. In

this dataset both Categorical and numerical columns

Hospital Readmission Risk Prediction Using Ensemble Learning

823

are included in the dataset. scalable modeling was

made possible by encoding categorical classiﬁcations

and normalizing the numerical features for scale con-

sistency. As the result, the dataset serves as a funda-

mental source for traing and testing ML models aimed

to forecast the possibility that more patients will re-

quire readmission.

4.2 Data Preprocessing

Cleaning the data begins with handling missing

values. The absence of data in a record, whether

intentional or not, is referred to as missing values.

Data inconsistencies alter algorithm performance and

Put in danger data integrity. Therefore, addressing

issues like bias in data becomes the second cleansing

stage. The last stage of feature optimization, which

in this case mainly comprises lowering the number of

unique values for categorical variables, is carried out

after the data has been cleaned for missing values and

other causes of bias. The various feature engineering

(Bahrami, ) steps feature creation, feature encoding,

outlier removal, feature selection are used. In fact,

while certain feature engineering processes depend

on the data and business knowledge, others such

as variable encoding, take into account what future

algorithms need to be used.

The proposed work demonstrates the training and

testing of ML models, that are MLP, XGBoost,

CatBoost for predicting the risk of readmission for

patients. To accurately assess model performance,

the dataset is separated into training and testing.

While the training set (80%) was utilized to train

the models, the testing set (20%) was reserved for

validation. In order to categorize readmission’s

according to the Excess Readmission Ratio threshold

( >1 for excess readmission’s), these are evaluated

using preprocessed data. several input features which

includes encoded category data and pre-processed

patient data, are used to get the expected result.

The MLP model is trained using regularization, early

stopping, and a single hidden layer to avoid over-

ﬁtting. 400 estimators, a learning rate of 0.01 and

regularization to lower model complexity are used to

train XGBoost. CatBoost contains 400 estimators,

regularization, and a learning rate of 0.01 and was

intended for categorical data. In an ensemble tech-

nique, the predictions produced by each model are

merged with those from the training and test sets. To

increase overall prediction accuracy and generaliza-

tion, the weighted ensemble approach was used to in-

crease the predictive accuracy of the hospital readmis-

sion risk prediction model. With weights of 0.2, 0.35,

and 0.35, respectively, the predictions of three mod-

els MLP, XGBoost, and CatBoost were combined to

make use of each model’s strengths.

[1] Training data (X

train

), Testing data

test

) Final predictions P

ensemble

and

evaluation metrics

Preprocessing: if categorical features exist

then

end

ncode them else

end

kip encoding if missing values exist then

end

andle them else

end

ontinue Scale numerical features as needed

Train Base Models: for each

m ∈ {CatBoost, XGBoost, MLP} do

end

m is XGBoost Tune hyperparameters else

end

se defaults Train m on (X

train

)

Generate Predictions: for each

m ∈ {CatBoost, XGBoost, MLP} do

end

ompute P

= m.predict(X

test

)

Deﬁne Weights: if XGBoost performs best

then

end

ssign higher w

MLP performs best Assign

higher w

else

end

ssign equal weights

Compute Ensemble:

ensemble

← w

· P

MLP

+ w

· P

XGB

+ w

· P

Cat

Evaluate and Return: if F1-score ≥ 0.8

then

end

roceed to deployment else

end

etrain models return P

ensemble

and metrics

Algorithm 1: Ensemble Learning Workﬂow

INCOFT 2025 - International Conference on Futuristic Technology

824

Table 1: Comparison of model accuracy for Hospital Read-

mission Prediction

Model Train Accuracy Test Accuracy

MLP 83.32% 82.69%

XGBoost 85.55% 84.43%

CatBoost 86.71% 86.60%

Ensemble 87.14% 87.08%

The output is the result of integrating the predic-

tions of individual models and an ensemble model.

The performance matrics includes accuracy, preci-

sion, recall, and F1-score are calculated for every

model. Table 1 show the comparison accuracy of train

and test of each model and Table 2 shows the perfor-

mance metrics of every model.

Table 2: Model Performance Metrics for Hospital Readmis-

sion Prediction

Model Accuracy Precision Recall F1-score

MLP 0.826897 0.822383 0.826897 0.821813

XGBoost 0.854328 0.859647 0.854328 0.844663

CatBoost 0.866045 0.869467 0.866045 0.858756

Ensemble 0.870839 0.874830 0.870839 0.863865

As shown in above Table 2, the Ensemble has

strong metrics on all performance, as Ensemble

model combines the predictions of each individual al-

gorithm, it has highest test accuracy of 87.08%.

Figure 5: ROC Curve graph of Models.

The ROC curve in Figure 5 analysis performed to

compare the performance of 4 models in prediction

of hospital readmission.With AUC value of 0.93 for

XGBoost and Ensemble, and 0.94 for CatBoost is a

best individual model in this case, outperforming XG-

Boost, a comparison of the AUC values shows that

the Ensemble model, XGBoost, CatBoost all have ex-

cellent predictive ability in identifying readmitted pa-

tients. While the MLP model achieved a moderate

AUC value of 0.88 but not as effective as other mod-

els.

5 CONCLUSION AND FUTURE

WORK

The study aimed to identify the optimal approach

for predicting hospital readmissions using machine

learning models. MLP, XGBoost, and CatBoost were

used to train models predicting readmission risk

based on the dataset features. XGBoost and CatBoost

outperformed MLP, with AUC scores of 0.93 and

0.94, while MLP with an AUC of 0.88. The ensemble

model, combining all three algorithms, achieved

an accuracy of 87.08%. These results demonstrate

that these algorithms can accurately predict hospital

readmissions. The study provides a foundation for

possible future developments in hospital readmission

prediction. Future work could focus on hyperparame-

ter tuning, advanced ensemble methods like stacking,

and incorporating additional data, such as medication

history, and treatment information, to further improve

performance.

REFERENCES

Bahrami, P. Using Ensemble Deep Learning and Feature

Engineering Approaches to Classify Hospital Read-

mitted Heart Disease Patients. PhD thesis, Toronto

Metropolitan University.

Bleich, J., Cole, B., Kapelner, A., Baillie, C. A., Gupta, R.,

Hanish, A., Calgua, E., Umscheid, C. A., and Berk,

R. (2021). Using random forests with asymmetric

costs to predict hospital readmissions. medRxiv, pages

2021–03.

Chen, R., Zhang, S., Li, J., Guo, D., Zhang, W., Wang, X.,

Tian, D., Qu, Z., and Wang, X. (2023). A study on

predicting the length of hospital stay for chinese pa-

tients with ischemic stroke based on the xgboost algo-

rithm. BMC medical informatics and decision making,

23(1):49.

Chopra, C., Sinha, S., Jaroli, S., Shukla, A., and Mahesh-

wari, S. (2017). Recurrent neural networks with non-

sequential data to predict hospital readmission of di-

abetic patients. In proceedings of the 2017 Inter-

national Conference on Computational Biology and

Bioinformatics, pages 18–23.

Hidayaturrohman, Q. A. and Hanada, E. (2024). Impact

of data pre-processing techniques on xgboost model

performance for predicting all-cause readmission and

mortality among patients with heart failure. BioMed-

Informatics, 4(4):2201–2212.

Kahn III, C. N., Rhodes, K., Pal, S., McBride, T. J., May, D.,

DaVanzo, J. E., and Dobson, A. (2023). Cms hospital

value-based programs: Reﬁnements are needed to re-

duce health disparities and improve outcomes: Study

Hospital Readmission Risk Prediction Using Ensemble Learning

825

examines cms hospital value-based programs’ effec-

tiveness in addressing health disparities and outcomes.

Health Affairs, 42(7):928–936.

Kalusivalingam, A. K., Sharma, A., Patel, N., and Singh, V.

(2012). Enhancing hospital readmission rate predic-

tions using random forest and gradient boosting algo-

rithms. International Journal of AI and ML, 1(2).

Leonard, G., South, C., Balentine, C., Porembka, M., Man-

sour, J., Wang, S., Yopp, A., Polanco, P., Zeh, H.,

and Augustine, M. (2022). Machine learning im-

proves prediction over logistic regression on resected

colon cancer patients. Journal of Surgical Research,

275:181–193.

Lopez, K., Li, H., Lipkin-Moore, Z., Kay, S., Rajeevan,

H., Davis, J. L., Wilson, F. P., Rochester, C. L., and

Gomez, J. L. (2023). Deep learning prediction of hos-

pital readmissions for asthma and copd. Respiratory

Research, 24(1):311.

Mahajan, S. M. and Ghani, R. (2019). Using ensemble ma-

chine learning methods for predicting risk of readmis-

sion for heart failure. In MEDINFO 2019: Health and

Wellbeing e-Networks for All, pages 243–247. IOS

Press.

Mienye, I. D. and Sun, Y. (2022). A survey of ensem-

ble learning: Concepts, algorithms, applications, and

prospects. IEEE Access, 10:99129–99149.

Quan, X. and Gopukumar, D. (2023). Use of claims data to

predict the inpatient length of stay among us stroke pa-

tients. Informatics in Medicine Unlocked, 42:101337.

Rao, T. S. S. and Battula, B. P. (2019). A frame work

for hospital readmission based on deep learning ap-

proach and naive bayes classiﬁcation model. Rev.

d’Intelligence Artif., 33(1):67–74.

Rizinde, T., Ngaruye, I., and Cahill, N. (2024). Machine

learning algorithms for predicting hospital readmis-

sion and mortality rates in patients with heart failure.

African Journal of Applied Research, 10(1):316–338.

Safaei, N., Safaei, B., Seyedekrami, S., Talaﬁdaryani, M.,

Masoud, A., Wang, S., Li, Q., and Moqri, M. (2022).

E-catboost: An efﬁcient machine learning framework

for predicting icu mortality using the eicu collabora-

tive research database. Plos one, 17(5):e0262895.

Slezak, J., Butler, L., and Akbilgic, O. (2021). The role

of frailty index in predicting readmission risk follow-

ing total joint replacement using light gradient boost-

ing machines. Informatics in Medicine Unlocked,

25:100657.

Teo, K., Yong, C. W., Chuah, J. H., Hum, Y. C., Tee,

Y. K., Xia, K., and Lai, K. W. (2023). Current

trends in readmission prediction: an overview of ap-

proaches. Arabian journal for science and engineer-

ing, 48(8):11117–11134.

Ti’jay Goudjerkan, M. J. Predicting 30-day hospital read-

mission for diabetes patients using multilayer percep-

tron.

Turgeman, L. and May, J. H. (2016). A mixed-ensemble

model for hospital readmission. Artiﬁcial intelligence

in medicine, 72:72–82.

Wang, T. and Paschalidis, I. C. (2019). Prescriptive cluster-

dependent support vector machines with an applica-

tion to reducing hospital readmissions. In 2019 18th

European Control Conference (ECC), pages 1182–

1187. IEEE.

Yu, K. and Xie, X. (2019). Predicting hospital readmis-

sion: a joint ensemble-learning model. IEEE journal

of biomedical and health informatics, 24(2):447–456.

Zhou, H., Ngune, I., Albrecht, M. A., and Della, P. R.

(2023). Risk factors associated with 30-day un-

planned hospital readmission for patients with mental

illness. International Journal of Mental Health Nurs-

ing, 32(1):30–53.

INCOFT 2025 - International Conference on Futuristic Technology

826