Enhancing heart disease classification based on greylag goose optimization algorithm and long short-term memory

Abstract

Heart disease is a category of various conditions that affect the heart, which includes multiple diseases that influence its structure and operation. Such conditions may consist of coronary artery disease, which is characterized by the narrowing or clotting of the arteries that supply blood to the heart muscle, with the resulting threat of heart attacks. Heart rhythm disorders (arrhythmias), heart valve problems, congenital heart defects present at birth, and heart muscle disorders (cardiomyopathies) are other types of heart disease. The objective of this work is to introduce the Greylag Goose Optimization (GGO) algorithm, which seeks to improve the accuracy of heart disease classification. GGO algorithm's binary format is specifically intended to choose the most effective set of features that can improve classification accuracy when compared to six other binary optimization algorithms. The bGGO algorithm is the most effective optimization algorithm for selecting the optimal features to enhance classification accuracy. The classification phase utilizes many classifiers, the findings indicated that the Long Short-Term Memory (LSTM) emerged as the most effective classifier, achieving an accuracy rate of 91.79%. The hyperparameter of the LSTM model is tuned using GGO, and the outcome is compared to six alternative optimizers. The GGO with LSTM model obtained the highest performance, with an accuracy rate of 99.58%. The statistical analysis employed the Wilcoxon signed-rank test and ANOVA to assess the feature selection and classification outcomes. Furthermore, a set of visual representations of the results was provided to confirm the robustness and effectiveness of the proposed hybrid approach (GGO + LSTM).

Keywords: Feature selection; Heart disease classification; LSTM; Optimization; bGGO.

Fig. 1

Pathological heart rhythm.

Algorithm 1

GGO Algorithm.

Algorithm 2

bGGO Algorithm.

Fig. 2

The proposed heart disease classification framework.

Fig. 3

Histogram plot for each feature in the dataset.

Fig. 4

Correlation matrix between features in the dataset.

Fig. 5

The average error of the results acquired using bGGO, the proposed feature selection technique.

Fig. 6

Analysis plots of the obtained outcomes based on bGGO, the proposed feature selection technique.

Fig. 7

Assessing the accuracy of the GGO + LSTM approach and optimization algorithms using the LSTM model, considering the objective function.

Fig. 8

Histograms of the accuracy results achieved by the GGO + LSTM approach, as well as alternative combinations of optimization techniques with LSTM models.

Fig. 9

Analysis plots of the results of the proposed GGO + LSTM and other algorithms.

Fig. 10

Regression plot: accuracy vs. F-score for the proposed GGO + LSTM approach and other algorithms.

Fig. 11

KDE plot of accuracy for the proposed GGO + LSTM approach and other algorithms.

Fig. 12

Sensitivity (TRP) by model for the proposed GGO + LSTM approach and other algorithms.

Fig. 13

Specificity (TNP) by model for the proposed GGO + LSTM approach and other algorithms.

Fig. 14

Boxplots for model metrics for the proposed GGO + LSTM approach and other algorithms.

Fig. 15

Pairplot with regression lines for the proposed GGO + LSTM approach and other algorithms.