Predicting COVID-19 patient recovery or mortality using deep neural decision tree and forest
Mohammad Dehghani1, Mohadeseh Zarei Ghobadi2, Mobin Mohammadi3
1School of Electrical and Computer Engineering, University of Tehran, Tehran, Iran. dehghani.mohammad@ut.ac.ir.
BMC Research Notes
|October 28, 2025
Summary
Deep neural decision forests accurately predict COVID-19 patient mortality using clinical data. This machine learning approach offers a reliable tool for identifying high-risk individuals, improving resource allocation in healthcare settings.
Area of Science:
- Medical Informatics
- Machine Learning in Healthcare
- Epidemiology
Background:
- Effective identification of high-risk patients is vital for emergency physicians, especially during global health crises like COVID-19.
- Resource allocation in healthcare systems, particularly in underserved regions, necessitates accurate mortality prediction.
- COVID-19 has highlighted the need for advanced tools to assess disease severity and patient outcomes.
Purpose of the Study:
- To evaluate the predictive performance of deep neural decision forests and deep neural decision trees for COVID-19 patient mortality.
- To compare the effectiveness of various machine learning and deep learning models in predicting mortality using diverse patient data.
- To assess the utility of clinical data, demographics, health indicators, and occupational factors in mortality prediction.
Main Methods:
- Utilized a stratified sampling method to partition a dataset of COVID-19 patients.
- Employed nine machine learning and deep learning methods, including deep neural decision forests and trees, to build predictive models.
- Focused on analyzing patient data encompassing diagnosis, demographics, health indicators, and occupational risk factors.
Main Results:
- The deep neural decision forest model demonstrated superior performance in predicting patient mortality.
- Using only clinical data, the deep neural decision forest achieved an accuracy of 80.7%, recall of 80.7%, precision of 75.7%, and F1-score of 74.8%.
- This model integrates deep neural networks' representational power with decision forests' interpretability, outperforming other methods without requiring imaging or laboratory tests.
Conclusions:
- Deep neural decision forests are effective and reliable predictors of mortality among COVID-19 patients.
- The model's ability to perform well using solely clinical data enhances its applicability in resource-limited settings.
- This approach offers an interpretable and high-performing solution for COVID-19 mortality risk stratification.
