Page 148 - EJMO-9-1

Eurasian Journal of Medicine and
Oncology
Machine learning insights into heart failure outcomes

predict certain instances of death events. Overall, while that the SVM model’s performance may be suboptimal for
the linear regression model showed promise in predicting this particular prediction task, potentially due to the data
death events among HF patients, there is potential for distribution or parameter settings. Further investigation
improvement, especially in minimizing false positive and and optimization of the SVM model, such as fine-tuning
false negative predictions. Further refinement of the model hyperparameters or exploring alternative kernel functions,
through feature selection, hyperparameter tuning, and may be necessary to improve its predictive accuracy and
potentially exploring alternative algorithms may enhance reliability. Overall, while the SVM model demonstrates
its predictive performance and clinical utility. In addition, high specificity in identifying true negative cases, its
the findings from the confusion matrix underscore the limited sensitivity in detecting positive cases highlights the
importance of evaluating model performance beyond need for continued refinement and evaluation to enhance
simple accuracy metrics, as it provides a more nuanced its clinical utility. The findings from the confusion matrix
understanding of the model’s strengths and limitations in underscore the importance of assessing both true positive
clinical decision-making. and false negative rates in interpreting model performance
The confusion matrix for the random forest model offers and guiding decision-making in clinical settings.
insights into its performance in predicting death events The confusion matrix for the GBM model provides
among HF patients. The model achieved a considerable insights into its performance in predicting death events
number of true positives, accurately identifying cases among HF patients. The model achieved a substantial
where death events occurred. However, it also produced number of true positives, accurately identifying cases
some false positives, incorrectly predicting positive where death events occurred. In addition, it exhibited a
outcomes where no death events occurred. Furthermore, high number of true negatives, correctly identifying cases
while the model demonstrated a high number of true where no death events occurred. However, the model
negatives by correctly identifying cases where no death also produced several false positive and had a notable
events occurred, it also had a notable number of false number of false negatives. While the false negative rate
negatives, failing to predict certain instances of death was lower compared to the false positive rate, it indicates
events. Although the random forest model performed well potential areas for improvement in the model’s sensitivity
overall, the presence of false positives and false negatives to positive cases. Overall, the GBM model demonstrates
suggests potential areas for improvement, such as refining promise as a predictive tool for identifying individuals
the model’s hyperparameters or incorporating additional at risk of death events in HF patients. Its performance,
features, to enhance predictive accuracy. Overall, the characterized by a balance between true positive and true
random forest model shows promise as a predictive tool negative predictions, suggests its potential utility in clinical
for identifying individuals at risk of death events in HF risk stratification. However, efforts to reduce false positive
patients. However, further optimization and validation are and false negative predictions through model optimization
necessary to ensure its reliability and generalizability in and feature engineering may further enhance its predictive
clinical practice. The findings from the confusion matrix accuracy and clinical applicability. The findings from the
highlight the importance of evaluating model performance confusion matrix underscore the importance of evaluating
comprehensively and considering both true and false model performance comprehensively and considering
predictions in interpreting results and guiding clinical both true and false predictions in interpreting results and
decision-making. guiding clinical decision-making. Continued refinement
The confusion matrix for the SVM model reveals and validation of predictive models are essential to ensure
its performance in predicting death events among HF their reliability and effectiveness in real-world healthcare
patients. Interestingly, the SVM model did not predict settings.
any positive outcomes for the actual positive cases, This study’s findings align with prior research,
resulting in a true positive count of 0. It correctly identified indicating that confusion matrices provide valuable
all negative outcomes, leading to a high count of true insights into the strengths and limitations of predictive
negatives. However, the inability to predict any positive models by categorizing predictions as true positives, false
outcomes for actual positive cases resulted in a high count positives, true negatives, and false negatives. For instance,
19
of false negatives, indicating instances where the model our observation of high true positive rates and relatively
failed to identify patients at risk of death events. While low false positive rates in certain models is consistent
the absence of false positives is desirable, the high rate with studies highlighting the importance of sensitivity
of false negatives suggests that the SVM model may lack and specificity in clinical decision-making. However,
20
sensitivity in detecting positive cases. The results suggest our identification of false negative instances underscores

Volume 9 Issue 1 (2025) 140 doi: 10.36922/ejmo.6583