Page 59 - AIH-1-1

Artificial Intelligence in Health AI model for cardiovascular disease prediction

TP + TN
Accuracy = (V)
+
+TP +TN FP FN
TP
Precision = (VI)
TP + FP

1 n 2
MSE = ( ∑ S −S ' ) (VII)
n t t
= t1
where TP is true positive that correctly classified positive
cases; TN is true negative that correctly classified negative
Figure 15. Performance analysis of support vector machine algorithms cases; FP is false positive that incorrectly classified positive
on datasets of different sizes. cases; FN is false negative that incorrectly classified negative
cases; MSE is mean squared error; n is classifications; S is
(t)
actual classification; and S’ is predicted classification.
(t)
5. Conclusion
Various classification techniques and a comprehensive
understanding of the hidden correlations between
attributes that play pivotal role in CVD are instrumental
for cost-effective, automatic, and early prediction of the
disease to reduce the mortality rate. This study worked
on different CVD attributes from patients using ANN,
ANN-GA, K-means, KNN, SVM, and DT in a MATLAB
environment. Given the diversity in the attributes and
dataset number, GA was employed for the selection of
Figure 16. Performance analysis of decision tree algorithms on datasets correlated attributes that contribute to CVD. The purpose
of different sizes.
of this work is to shed light on different classifiers with
66.70%, 0.3300, 76.00%, 55.00%, and 67.90%, respectively. a better predictive ability (precision) since wrong and
For 1000 datasets, the accuracy, MSE, sensitivity, specificity, late diagnosis may lead to death. The performance of
and precision were 76.70%, 0.3400, 94.10%, 53.80%, and the classifiers was evaluated in terms of accuracy, MSE,
72.70%, respectively. For 50 datasets, the accuracy, MSE, sensitivity, specificity, and precision. Based on the results,
sensitivity, specificity, and precision were 50.00%, 0.5000, the ANN model combined with GA performs better with
85.30%, 46.00%, and 50.00%, respectively. Using the DT an accuracy of 86.4% as compared to SVM at 84.0%,
algorithm, the classification accuracy, sensitivity, and K-means at 59.6%, KNN at 79.0%, and DT at 77.8%. Thus,
precision decreased as the number of datasets reduced to the ANN-GA model is therefore recommended for CVD
1500 and then increased at 1000 datasets before decreasing diagnosis and prediction. This research shows that better
again at 500 datasets. MSE was increased as the number of accuracy is obtained when a larger number of datasets
datasets was reduced from 2000 to 500. On the other hand, are used. Future research work should focus on expert
the specificity increased at first as the number of dataset system development for the CVD prediction, diagnosis,
reduced from 2000 to 1500 before a steady decrease with and prescription of drugs. Furthermore, more robust AI
the reduction in the number of datasets. and optimization algorithms should be developed for the
optimal performance of CVD prediction and diagnosis.
The formulas for determining sensitivity, specificity,
accuracy, precision, and MSE are given in the Equations Acknowledgments
III to VII:
The authors appreciate the management of Federal
TP
Sensitivity = (III) University of Technology, Minna, Nigeria, for the given
TP + FN access to the AI and Embedded System Laboratory in
Computer Engineering Department, where thorough
TN
Specificity = (IV) research investigation and coding were carried out. The
TN + FP authors also extend their sincere appreciation to the

Volume 1 Issue 1 (2024) 53 https://doi.org/10.36922/aih.1746