A self-adjusting ant colony clustering algorithm for ECG arrhythmia classification based on a correction mechanism.

• In this paper, ECG arrhythmia classification was achieved without distinguishing subjects. Not only the classification within subjects, but also the classification among subjects. • This paper improved the ant colony clustering algorithm by introducing a correction mechanism to correct outlier points to improve model classification accuracy; introducing an increased flow rate ρ to improve the speed and stability of model convergence; and realising that the next transfer target is selected by a true self-adjusting transfer method where the transfer probability is dynamically adjusted according to pheromone concentration and path distance. • This paper compares different models with and without the correction mechanism, and experiments show that the number of classification errors is reduced by an order of magnitude (101). • Compared with the experimental models, the classification accuracy of the proposed method is improved by 0.2% ∼ 16.6%. Compared with other current studies, the classification accuracy of the proposed method is improved by 0.65% ∼ 7.5%. As a representative type of cardiovascular disease, persistent arrhythmias can often become life-threatening. In recent years, machine learning-based ECG arrhythmia classification aided methods have been effective in assisting physicians with their diagnosis, but these methods have problems such as complex model structures, poor feature perception ability, and low classification accuracy. In this paper, a self-adjusting ant colony clustering algorithm for ECG arrhythmia classification based on a correction mechanism is proposed. This method does not distinguish between subjects when establishing the dataset in order to reduce the effect of differences in ECG signal features between individuals, thus improving the robustness of the model. When the classification is achieved, a correction mechanism is introduced to correct outliers caused by the accumulation of errors in the classification process in order to improve the classification accuracy of the model. According to the principle that the flow rate of gas can be increased under the convergence channel, a dynamically updated pheromone volatilization coefficient ρ, namely the increased flow rate ρ, is introduced to help the model converge more stably and faster. As the ants move, the next transfer target is selected by a truly self-adjusting transfer method, and the transfer probability is dynamically adjusted according to the pheromone concentration and the path distance. Based on the MIT-BIH arrhythmia dataset, the new algorithm achieved classification of five heart rhythm types, with an overall accuracy of 99.00%. Compared to other experimental models, the classification accuracy of the proposed method represents a 0.2% to 16.6% improvement, and compared to other current studies, the classification accuracy of the proposed method is 0.65% to 7.5% better. This paper addresses the shortcomings of ECG arrhythmia classification methods based on feature engineering, traditional machine learning and deep learning, and presents a self-adjusting ant colony clustering algorithm for ECG arrhythmia classification based on a correction mechanism. Experiments demonstrate the superiority of the proposed method compared to basic models as well as those with improved partial structures. Furthermore, the proposed method achieves very high classification accuracy with a simple structure and fewer iterations than other current methods. [ABSTRACT FROM AUTHOR]