Application of Extreme Learning Machine Method With Particle Swarm Optimization to Classify of Heart Disease

Coronary heart disease is the condition where the heart’s blood supply is blocked. Heart disease is the leading cause of death worldwide. Various risk factors contribute to heart disease, including smoking, unhealthy lifestyle, high cholesterol


INTRODUCTION
Coronary heart disease, or cardiovascular disease, occurs when blood flow to the heart muscle is narrowed or blocked by blood vessels that cause a heart attack, accompanied by chest pain or stroke.Heart disease is a disease that causes many deaths around the world.According to the World Health Organization (WHO), more than 17 million people have died from heart disease.Smoking, unhealthy lifestyle, high cholesterol, and hypertension are just a few risk factors that contribute to heart disease [1][2] [3].
In an effort to prevent an increase in the number of deaths from heart disease, it is possible to predict heart disease in humans.There are many technologies that can be used to manage data that can help determine whether a person has a risk of heart disease or not.Data mining is one of the most widely used technologies.Data mining is the process of gathering important information from large amounts of data.Data Mining is used for the process of collecting important information related to informatics; data mining is also used in the health section for the prediction of diseases, one of which uses classification methods.Classification is a grouping of data in which the data has a label class or a target.There are several methods of classification, including Extreme Machine learning (ELM).This method is found in the nerve network [4] [5] [6].
Extreme Learning Machine (ELM) is a feedforward artificial neural network which has one hidden layer called Single Hidden Layer Feedforward Neural Networks (SLFNs).Feedforward aims to recognize patterns in data so that the identification process has accurate results.Feedforward has the best weight search process by searching from an activation function.In the ELM algorithm the hidden node learning parameters including input weights and input bias can be set randomly, and for network output weights can be determined analytically with a simple general inverse operation [7] Based on research conducted by [8], Extreme Learning Machine is a method of simulated neural networks that has the advantage that its training process does not require reverse propagation, so the training process can be faster because it does not need to update the weight and bias values of each epoch.ELM methods can be used to detect the presence of brain tumors by performing classifications on MRI imaging data.There are two types of tumors: tumors and non-tumors.Previous research was conducted [8] [9] on the comparison of ELM and K-NN algorithms.Test results of both algorithms for the classification of heart disease using the ELM algorithm resulted in the best performance with an average accuracy value of 93.33%, while the K-NN algorithm obtained a lower average accuracy value of 83,52%.This is because the performance of the K-NN algorithm is heavily influenced by the value of k. High value of k results in the K-NN not performing at its maximum.Therefore, the application of ELM methods is expected to produce more effective predictions for the diagnosis of heart disease.
Another study using ELM was carried out by [11] on the classification of tuberculosis diseases (TB).ELM yielded an accuracy of 99.33% using optimal parameters for the tests already obtained.The optimum number of hidden neurons was 20, the binary sigmoid activation function was the optimum function in this research, and the percentage of training and testing data was optimum at 70%:30%.In addition, research conducted by [12] reported that in addition to having the advantages of ELM, it also had a weakness, namely that the number of hidden nodes was determined by trial and error, so it was not possible to know the correct number of hidden nodes to get the correct results using the ELM method.In addition to the problem of hidden nodes in making the selection of input weight, if the bias in the ELM method values is chosen randomly, then this can result in the calculation being less than maximum.
A research was conducted by [13] on the assessment of apartments using the SVM method based on PSO.The research resulted in an accuracy of 82.04%, while using SVM alone produced an accuracy of 79.20%.These results show that the Particle Swarm optimization-based Vector Machine support method is a fairly good method for data classification.
Therefore, based on the abovementioned problem, the current research was conducted on the classification of heart disease utilizing the Extreme Learning Machine (ELM) approach with Particle Swarm Optimization (PSO) optimization with rapid learning to result in optimal outcomes.

METHODS
The flow of the research is depicted in Figure 1.

Data Collection
This research used heart disease data obtained from the publicly available Kaggle Datasets website.The data used in this research had 14 attributes and consisted of 303 data.The data used had two values: 0 for non-heart disease conditions and 1 for heart disease conditions.This is shown in Table 1.

Preprocessing
This phase used data normalization using the Min-Max Normalization method.Min-Max Normalization is a method of normalization that involves performing a linear transformation of the original data to produce a balance of comparative values between the data before and after the process.This method can be done using the formula in Equation 1.

Designing the Model
The dataset was divided into two parts using the split data technique, consisting of 70% training data and 30% test data, after going through the preprocessing stage.This is based on a previous research, namely the classification of breast cancer using the Naïve Bayes algorithm.Other research conducted by [11] on the classification of tuberculosis diseases using the Extreme Machine Learning Method produced the highest average accuracy at 70% and 30% percentages.This research was divided into two classification models: Extreme Learning Machine classification and Extreme Machine Learning classification with Particle Swarm optimization.

Evaluation
Evaluation is a stage to measure the performance of the data mining modelling.In the evaluation stage, accuracy was determined using confusion matrix.This is shown in Table 2 and Equation 2. The classification accuracy can be calculated from the confusion matrix as the sum of correct cells in the table (true positives and true negatives) divided by all cells in the table.

Preprocessing
The stage of data preprocessing in this research was to transform smaller distance values (0-1) using the min-max normalization method.Normalization was used for data standardization so that both input range and output range are in the same range between 0 and 1.This is shown in Figures 2 and 3.

Testing of Populations
The test aimed to identify the number of populations that can provide optimal evaluation values for cases of diagnosis of heart disease using the ELM method optimized using PSO.In this test, an inertia weight (w) of 0.5 was used with c1 and c2 which had a value of 1, the number of iterations was 2, and the number of hidden nodes for ELM was 5.The division of training data and test data was 70%:30%.The size of population variations started at 100 and multiplies to 500.For each variation, the number of populations was tested five times, and then the average for each evaluation was obtained.The results of the test are provided in Table 3.Based on Table 3 above, optimal evaluation resulted for the population of 300 with an average of 83.30%.

Testing of the Number of Hidden Node
This test aimed to find out how many hidden nodes could produce an optimal evaluation.This test employed the parameter values of the optimal evaluation results that have been performed previously, namely the Population test with a total population of 300, an inertia weight of 0,5, and constants 1 (c1) and 2 (c2) of 1.This test was performed five times with the number of hidden nodes being 3, 5, 7, 9, and 11, and then the average for each evaluation was obtained.The number of hidden node testing is given in Table 4.

Testing of the number of iterations
Testing based on the number of iterations aimed to obtain the number of iterations that can produce the best evaluation value in of heart disease cases.In this test, the number of iterations used was 3.5, 10, 15, and 20.In each variation, the number of iterations was performed five times.The number of hidden nodes was 9, the value of inertia weight was 0.5, and the constants 1 (c1) and 2 (c2) are 1.The number of iterations can be seen in Table 5.The decrease in average fitness was consistent.It can be concluded from Table 5 that the average value of fitness is influenced by the number of iterations.The more the iterations, the lower the value of fitness.The best fitness score is of iteration 3 with an average accuracy of 83.64%.

ELM testing
This test aimed to find out the evaluation value of the application of the ELM method with PSO-ELM in conducting classifications in the diagnosis of heart disease.The ELM test without PSO used nine hidden nodes.The program was ran 5 times.Then, an average of the evaluation score was obtained.Table 6 shows the average of the evaluation results on the application of ELM for the diagnosis of heart disease.From Table 6, it can be seen that the average evaluation score is 57,32%.

ELM testing with PSO
This test aimed to find out the best evaluation value in cases of heart disease using an ELM optimized with PSO.This test used the parameter value of the optimal evaluation result that has been performed previously: the number of hidden nodes of 9 pieces with the number of iterations of 3. The inertia weight was 0,5, constants 1 (c1) and 2 (c2) are 1.This test was also done five times, and the average of the evaluation value was obtained.The ELM testing with PSO can be seen in Table 7. From Table 7, it can be seen that the result of ELM testing using the optimal parameters was established.The average accuracy value is 83.74%.

CONCLUSIONS
Based on the research that has been conducted using Extreme Learning Machine and Particle Swarm Optimization for the classification of heart disease, it can be concluded that the evaluation results on the diagnosis process of coronary heart disease using the ELM method produced an accuracy value of 57,32%, while by performing the optimization of ELM parameters using the PSO method, the best parameter obtained from the test results was the number of hidden nodes of 9, the total population of 300, the best number of iterations of 3, the inertia weight of 0.5, and the constants 1 (c1) and 2 (c2) of 1.The accuracy of the parameter was 83.74%.It can be concluded that the PSO method used to obtain ELM parameters such as input weight and bias influenced the results of the evaluation on the case of classification for the diagnosis of heart disease using the ELM method.The addition of the PSO algorithm to the ELM classification aimed to obtain better classification results than the conventional ELM.This has been demonstrated in this research with conventional ELM accuracy results that are much smaller than the accuracy of ELM & PSO.The accuracy obtained from the conventional ELM algorithm was only 57.32%.For further development of the research, there are some suggestions from the author, among others, to use algorithms or other optimization methods to obtain optimal ELM parameters so that it can increase the evaluation value of the classification for the diagnosis of coronary heart disease.Oher methods of classification to find out which classification methods can provide better evaluation values in the case of a diagnosis of heart disease can also be used.

Figure 1
Figure 1 The Research Flow

1 =
exercise relative to rest X11 Slope The slope of the peak exercise segment: Nilai 1 : up sloping Nilai 2 : flat Nilai 3 : down sloping X12 Ca Number of major vessels colored by dluoroscopy that ranged between Patient who is subject to possible heart disease IJCCS ISSN (print): 1978-1520, ISSN (online): 2460-7258 ◼ Application of Extreme Learning Machine Method With Particle …( Adela Putri Ariyanti) 285 2.
): 1978-1520, ISSN (online): 2460-7258 IJCCS Vol.17, No. 3, July 2023 : 281 -290 286 The data collection process was done through Kaggle Datasets that are publicly available.The data consisted of 303 records, consisting of 165 records of ill class label attributes and 138 records of healthy class label attributes.

Figure 2 Figure 3
Figure 2 Data Before the Normalization

Table 2
Confusion Matrix

Table 3
Results of Population Testing

Table 4
Results of Testing the Number of Hidden Node

Table 5
Results of Testing the Number of Iterations

Table 6
Results of the ELM Test

Table 7
Results of ELM test with PSO