Human heartbeats 70 times per minute approximately. Three important electrical potentials of the heart are P Wave, QRS Wave, and T Wave. The raw ECG signal is given as input and the signal is smoothed by pre-processing after pre-processing the signals are passed to feature selection, here feature selection method is compared with the modified techniques and standard algorithms. Frequency (Hz) = 1/Time period (Seconds(s) hence F=1/T 1/500ms=1/0.5s=2Hz.2Hz is 120 beats/minute. 500 ms is the time period between peaks of an ECG of a normal healthy person. In this paper, modified techniques are employed with the standard algorithm and presented to the classifier to predict the accurate result. The raw ECG signal is given as input and the signal is smoothed by pre-processing after pre-processing the signals are passed to feature selection, here feature selection method is compared with the modified techniques and standard algorithms. The selected feature is presented to the classifier for accurate classification result.
Figure2. Proposed Methodology
DATABASE AND METHODOLOGY
The data are collected from the MIT BIH Arrhythmia database which contains 48 records recorded for 30 minutes with a sampling frequency hertz of 360 Hz. 1975, Massachusetts Institute of Technology’s laboratory Beth Israel Hospital at Boston’s was supported to produce the MIT-BIH Arrhythmia Database. To evaluate the Arrhythmia disease, Massachusetts Institute of Technology provided the MIT-BIH Arrhythmia Database, which was the first available set of standard data set for the Arrhythmia research. Since 1980 onward this database was available in 500 sites across the world, which paves the way to do research in finding and diagnosing Cardiac Arrhythmia. Among that, 60% of data sets were collected from inpatients and 40 % of data sets were collected from outpatients. The approach used in our paper is divided into the following aspects: Pre-processing, Feature selection, Training the classifier and evaluation of the result. In feature selection, three modified techniques are used to select the most relevant features. The modified PSO, BFO and BAT algorithms are employed to the feature selection and the selected relevant feature is presented to the modified SVM and ELM classifier and the result is presented.
The foremost objective of pre-processing is to transform the noise signal into a reliable format for further analysis and remove the noise present in the signal to make the data more close to relevant data. Noise immunity is a major asset of the electronic device. The important and the preliminary step in this ECG recording is to remove the noise associated with the data. If the signal or input data is pre-processed the noisy interactions of power line interface, baseline drifting, muscle contraction, generated by the equipment will be eliminated and produced the noiseless signal to feature selection. In this work, wavelet function is used to diagnose and de noise the signal. The input records are taken from the MIT BIH database, and the signals are loaded into the Mat lab wavelet function toolbox and denoised the signal as shown in the figure3. Sample record from MIT BIH database are shown in figure 4 (a) & (b)
Figure3: Preprocessing signal using Wavelet function
Figure 4 (a) Original signal (b) Noise Removed record. (Record number: 100 sample record)
The important objective of feature selection is to find the minimal feature from the original data and to obtain high accuracy result.If we have, a large number of data’s it is very difficult to identify the exact pattern. This feature selection technique is used to remove the redundant, misleading, noisy, irrelevant signal from the pre-processed data. This process helps to select the best feature and presented to the classifier for accurate classification which minimizes the cost of the system and this helps to identify the accuracy. In this work, modified techniques are given and compared with the previous work and the standard algorithm and show that the modified technique gives the best result for the arrhythmia disease classification. Generally, noise reduction is the main issue which is to be solved by extracting the preferred noiseless signal for processing the further work. In the proposed work of feature selection method the modified PSO algorithm, Modified BFO, and BAT algorithm are compared and presented in the preceding section.