Predicting Hospital Readmission with a Hybrid LSTM-CNN: An Evaluation of Deep Learning Techniques in Healthcare Analytics

Thekkekaripurath Krishnankutty, Thushar (2024) Predicting Hospital Readmission with a Hybrid LSTM-CNN: An Evaluation of Deep Learning Techniques in Healthcare Analytics. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (1MB) | Preview
Preview	PDF (Configuration Manual) Download (929kB) | Preview

Abstract

This paper establish that hospital readmission is a key issue in the provision of healthcare services since it contributes to higher costs, inadequate resource utilization and patient’s poor health status. Risk assessment of readmission is relevant to interventive approaches and overall optimization of patient care. This paper aims to analyze a model that combines LSTM and CNN for predicting readmissions, using multiple EHR data. As a result, the proposed model integrates the benefits of identifying temporal dependence using LSTMs and identifying temporal structural patterns using CNNs to enhance the predicted readmissions’ dependability.

The research starts by highlighting basic issues with existing prediction methods including inadequacy in dealing with temporal data, and poor accuracy with high-dimensional data. Therefore, a stepwise approach that includes data preprocessing, EDA, feature engineering, and model selection was used for the study. Hence, the proposed LSTM-CNN model was constructed, and its performance was compared with standalone LSTM and CNN models. The success of the model was measured using accuracy and precision, recalls, F-1 score and ROC-AUC.

The experiments suggest that the use of the hybrid architecture is better than the separate models – the accuracy is at 63%, the precision is at 63%, and the recall is 63% on the test data set. However, the presented work also presents difficulties, for example, in equal distribution of classes in a dataset or incomprehensible intricacies of Deep Learning. However, the results obtained from the proposed approach showed promising results in utilising both the structural and temporal data in predictive analytics in healthcare applications.

This research evidences the potential of incorporating sophisticated machine-learning models in radical healthcare systems. The proposed model will enable healthcare providers to properly identify patients who are most likely to be readmitted and, therefore, health resources can be put to better use, unnecessary admissions prevented, and patients’ lives improved. Additionally, this presents limitations like data imbalance, one hot encoding, computational time and cost, and methodologically presents ideas for future work inclusion of explainable AI, multimodal data, and better ways for tuning up hyperparameters.

Therefore, the use of a hybrid LSTM-CNN model is a promising prospect in the complex field of hospital readmission prediction. They may be labelled on historical grounds as descriptive or prescriptive approaches but their main merit may be located in their realism for implementation where serious healthcare delivery systems can be based on accurate data. This research is connected to the existing literature on the application of artificial intelligence in healthcare, including the demand for further advancement in and future enhancement of prediction algorithms.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Subhnil, Shubham UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science R Medicine > R Medicine (General) Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	05 Sep 2025 11:46
Last Modified:	05 Sep 2025 11:46
URI:	https://norma.ncirl.ie/id/eprint/8826

Actions (login required)

View Item