An Optimized Hybrid ARIMA-GARCH Model Application on RR Interval Time Series Prediction for Heart Disease

Abstract

Heart disease is one of the highest mortality rate diseases worldwide, with arrhythmias frequently serving as a trigger (such as cardiomyopathy) or a complication (such as coronary heart disease) for cardiovascular diseases. Therefore, it is crucial to monitor abnormalities in heart function through the early identification of deviations in heart rate variability (HRV). In modern medical systems, wearable real-time monitoring devices and artificial intelligence are commonly employed to generate electrocardiograms (ECGs) and analyze HRV data. The key to this application lies in making reasonable judgments of HRV data using data mining tools, including multiple linear regression, support vector machine, random forest, or long-short-term memory neural networks. However, these models fail to yield satisfactory results for cardiac rhythm monitoring. Consequently, the paper introduces an optimized hybrid ARIMA-GARCH model to enable heart disease detection and pathological diagnosis, playing a guiding role in personalized treatment and the tracking of the cardiovascular health status of monitored individuals. The proposed model combines data preprocessing using the one-sided Hodrick Prescott filter and parameter tuning based on partitioning-interpolation techniques and Fast Discrete Fourier Transform to fit and predict the RR interval time series. Experimental results indicate that our proposed model exhibits significant advantages in quantitative assessments compared to other models, as it effectively preserves the trend and accounts for high volatility in short-term forward prediction.