<?xml version="1.0" encoding="UTF-8"?>



<records>

  <record>
    <language>eng</language>
          <publisher>Oriental Scientific Publishing Company</publisher>
        <journalTitle>Biomedical and Pharmacology Journal</journalTitle>
          <issn>0974-6242</issn>
            <publicationDate>2026-11-04</publicationDate>
    
        <volume>19</volume>
        <issue>4</issue>

 
    <startPage></startPage>
    <endPage></endPage>

	    <publisherRecordId>73989</publisherRecordId>
    <documentType>article</documentType>
    <title language="eng">Deep Autoencoder–Driven Hybrid Machine Learning Framework for Short-Term Mortality Prediction in Heart Failure Patients</title>

    <authors>
	 


      <author>
       <name>Sandeep Kumar Sunori</name>

 
		
	<affiliationId>1</affiliationId>
      </author>
    

	 


      <author>
       <name>Govind Singh Jethi</name>


		
	<affiliationId>1</affiliationId>

      </author>
    

	 


      <author>
       <name>Pradeep Juneja</name>

		
	<affiliationId>1</affiliationId>
      </author>
    

	


	


	
    </authors>
    
	    <affiliationsList>
	    
		
		<affiliationName affiliationId="1">Department of ECE, Graphic Era Hill University, Bhimtal Campus, India</affiliationName>
    

		
		
		
		
		
	  </affiliationsList>






    <abstract language="eng">Heart failure is one of the leading causes of hospitalization and mortality globally, underscoring the requirement of clinically implementable risk-stratification tools to predict short-term mortality. Traditional statistical frameworks and single-algorithm machine learning techniques frequently fail in capturing the complex, non-linear and heterogeneous nature of interactions in clinical data, particularly with small sample size. In this paper, a 2-step deep hybrid learning model combining unsupervised deep feature engineering and advanced hybrid classification architectures will be proposed to improve short-term mortality prognostication in heart failure patients. The model has been tested on the Heart Failure Clinical Records dataset (N = 299 patients, 11 clinical features) using five-fold cross-validation. An autoencoder compresses the input features into a compact 3-dimensional latent representation. The mean squared reconstruction error of autoencoder is 6.1376. Three hybrid models, which are RF-SVM, LSTM-XGBoost, and LSTM-SVM, are then utilized with these features. Also, the Modified Category Net Reclassification Improvement (MCNRI) metric is used to determine the clinical usefulness. Findings indicate small betterments than baseline models. The most successful model (LSTM-SVM) had a moderate AUC 0.71 and an accuracy of 72.9 % and with a small improvement over the baseline models. Although LSTM-XGBoost demonstrated superior risk reclassification (MCNRI = +0.16), the sensitivity of all the models was low (0.458), which indicates that about half of genuine mortality cases were not detected, which is a significant drawback of using the method in the real-world clinical setting. These findings suggest that deep feature engineering has some potential but the intermediate predictive accuracy and low sensitivity, coupled with the limited sample size (N = 299) make it impossible to implement it in clinical practice at the moment. Validation on larger, multi-centre datasets is required before clinical translation. Greater validation on more multi-centers datasets is needed in order to validate generalizability.</abstract>

    <fullTextUrl format="html">https://biomedpharmajournal.org/vol19no4/deep-autoencoder-driven-hybrid-machine-learning-framework-for-short-term-mortality-prediction-in-heart-failure-patients/</fullTextUrl>

<keywords language="eng">

      
        <keyword>Autoencoder</keyword>
      

      
        <keyword> Clinical risk stratification</keyword>
      

      
        <keyword> Deep feature engineering</keyword>
      

      
        <keyword> Heart failure</keyword>
      

      
        <keyword> Hybrid machine learning</keyword>
      

      
        <keyword> LSTM (Long Short-Term Memory)</keyword>
      

      
        <keyword> MCNRI (Modified Category Net Reclassification Improvement)</keyword>
      

      
        <keyword> Support vector machine</keyword>
      

      
        <keyword> XGBoost (eXtreme Gradient Boosting)</keyword>
      
</keywords>
  </record>
</records>