A New Automatic Identification Method of Heart Failure Using Improved Support Vector Machine Based on Duality Optimization Technique

Currently, Heart failure disease is considered a multifaceted clinical disease affecting millions of people worldwide. Hospitals and cardiac centers rely heavily on ECG as a regular tool for evaluating and diagnosing Heart failure disease as an initial stage. The process of Heart failure disease identification from the ECG signal aims to reduce the time of the diagnostic process for patients with heart failure and to improve the outcomes of the detection process applied to these patients. The information acquired from the ECG signal simplifies the laboratory evaluation and other traditional diagnosis methods to evaluate and diagnose Heart failure disease. Unluckily, the problem of segmentation of the ECG signal is complicated because of similarities in time interval and amplitude between several ECG signal as well as the presence of noise in ECG signals. Most cardiologists use the ECG signal to identify Heart failure disease and their decision depends on the identification process, to determine whether surgery or medical treatment is required. This paper offers a new identification technique to overcome the current problems, such as overlapping in heart rate duration from the time interval from one PQRST wave to the next. In this study, the aim is to develop a new automatic method using improved support vector machine to diagnosis HFD from ECG signals. This is particularly relevant to ECG signals for the diagnosis of HFD as the first step to treatment and care of patients in general and specifically those with early heart disease to increase their overall survival. This paper outlines a hybrid approach of dual SVM and nonparametric algorithm to spot HFD in ECG signals leading to increase reliability and accuracy of identification and diagnosis of heart failure classes in the early stages using the proposed algorithm. The nonparametric algorithm is used to train SVM and its dual to get two models of SVM. The dual problem gives a different view that is better and sometimes simpler than the original problem. This feature is used to detect Heart failure disease in ECG signals by comparing the outputs of SVM model and those of dual SVM model. Experiments show that the hybrid approach produces good results, is more efficient and increases accuracy of Heart failure disease detection with an acceptable accuracy of 94.97% when compared with other algorithms to which the paper refers to. This is especially noted in patients with multiple diseases who were not initially identified as heart failure.