Page 51 - AIH-1-1

Artificial Intelligence in Health AI model for cardiovascular disease prediction

the synthetic minority oversampling technique (SMOTE) 3. Materials and methods
to enrich the data. Regarding model performance in terms
of discrimination, GBT and RF were the models with the 3.1. Implementation of ANN-GA for CVDs model
highest receiver operating characteristic curve of 97.8% In this research, a CVD prediction model was developed
and 97.7%, respectively. The discriminative powers of ET, using a combination of techniques based on the evolution
LR, and DNN were 96.8%, 96.4%, and 95.3%, respectively, theory – GA and ANN. This ANN-GA model was used
with the latter exhibiting the least discriminative power. In for the correlated attribute selection, training, and
terms of reliability, LR predictions were the least calibrated optimization of the CVD selected features to achieve
compared with the other four algorithms. Thus, despite better prediction accuracy. The ANN utilized a multilayer
increasing the training time, SMOTE was proven to further perceptron to receive the transmitted signal, process the
improve the model performance of LR, while algorithms
like GBT and DNN performed well with class-imbalanced signal (neurons), and compute the output of each neuron
data . for weighted using back-propagation to adjust the NN
[43]
model parameters for a reduced mean squared error
Khan et al. utilized a machine learning algorithm (MSE). While the GA is an optimization searching agent
for the accurate prediction and decision-making for in the evolution theory, that is used to optimize the output
CVD patient . Simple random sampling was used to performance of a classifier model using a supervised
[44]
select heart disease patients from the Khyber Teaching machine learning algorithm (SMLA). The function of GA
Hospital and Lady Reading Hospital, Pakistan. Some of evolution initiates the process from a randomly generated
the machine learning methods involved are DT, RF, and population for an individual which changes at every
LR. The performance of the proposed machine learning iteration in the loops to formulate an offspring called
algorithm was estimated using numerous conditions to [45]
recognize the most suitable machine learning algorithm generation The fitness (best chromosome) of every
in the class of models. The RF algorithm has the highest individual generated within the population was appraised,
accuracy of prediction, sensitivity, and recursive operative and the fittest was chosen (selection of chromosome) by
characteristic curve of 85.01%, 92.11%, and 87.73%, computing the probability for being selected from among
respectively, for CVD prediction. It also has the least the best chromosome in the population in proportion
specificity and misclassification errors of 43.48% and to the sum of the fitness among individuals as given
[46]
8.70%, respectively. The overall performance evaluation in Equation I . Nevertheless, the generation of the
result showed that the RF algorithm is the most appropriate individual population can be modified by crossover and
algorithm for CVD classification and prediction . mutation to form a newly evolved generation candidate
[44]
A summary of the authors’ focus, methods, and strengths as presented in Figure 1. The overview of the ANN-GA
is presented in Table 1. flowchart is presented in Figure 2.

Table 1. Summary of the related works in terms of the authors’ focus, methods, and strength
Authors’ focus Methods Performance accuracy
Proposed CVD prediction model [35] Using machine learning algorithms such as K-neighbor 95.62%, SVM 95.62%, DT 91.05%,
SVM, RF, Naive Bayes classifier, and LR RF 95.40%, and LR 95.59%
Performance of data mining classification Using three popular classification techniques, KNN 82.96%, SVM 85.18%, and ANN 73.33%
techniques for heart disease prediction [36] such as KNN, SVM, and ANN
Proposed machine learning algorithm for heart Using PSO and ACO The average accuracy for PSO and ACO is 99.65%
disease prediction and classification [38]
Machine learning approaches for predicting, SVR, multivariate adaptive regression splines, ANFIS 96.56%, and SVR 91.95%
classifying, and improving the diagnosis M5Tree model neural networks, adaptive
accuracy of CVDs [42] neuro-fuzzy, and KNN, and Naive Bayes classifiers
Nurturing clinicians for accurate prediction of LR, RF, ET, and GBT GBT 97.8%, RF 97.7%, ET 96.8%, LR 96.4%,
mortality among CVD patients [43]
Accurate prediction and decision-making for DT, RF, LR DT 85.01%, RF 92.11%, and LR 87.73%
CVD classification [44]
Abbreviations: ACO: Ant colony optimization; ANFIS: Adaptive neuro-fuzzy inference system; ANN: Artificial neural network; CVD: Cardiovascular
disease; DT: Decision tree; ET: Extra tree; GBT: Gradient boosting tree; KNN: K-nearest neighbor; LR: Logistic regression; PSO: Particle swarm
optimization; RF: Random forest; SVR: Support vector regression.

Volume 1 Issue 1 (2024) 45 https://doi.org/10.36922/aih.1746