Page 101 - The-5th-MCAIT2021-eProceeding

3. Methodology

Fig. 2. Flowchart of proposed research.

3.1 MIT-BIH Arrhythmia Dataset:
We used the well-known arrhythmia database from the Massachusetts Institute of Technology (MIT). The
MIT-BIH Arrhythmia Dataset is the original dataset and publicly available (Moody and Mark, 2005). The MIT-
BIH arrhythmia contains a total of 48 records, each lasting approximately 30 minutes, and obtained between
1975 and 1979 from a two-channel ambulatory device (Moody et al., 2001).

3.2 Data Preparation:
The method of cleaning and converting raw data prior to processing and analysis is known as data
preparation. Before processing, it is a critical stage that always includes reformatting data, making data
corrections, and merging data sets to refine data. Data preparation ensures data integrity, resulting in accurate
observations (Picon et al., 2019).

3.3 BiLSTM Implementation:

BiLSTM layers are then implemented. Specifically, five cases are considered where the number of
BiLSTM layers is varied from 1 to 5. In addition, out of 48 records, data of 36 patient's record was utilized by
training module, and 12 patient's record was used for validation. Also, a random approach was utilized to split
the data into training and validation modules.

3.4 Performance Evaluation:
The performance of the proposed BiLSTM model was measured in terms of accuracy, recall, precision,
and specificity. The four metrics are defined as follows:
 Accuracy can be characterized as "the degree to which a measurement's outcome conforms to the
right value or a norm," and it refers to how similar measurement is to its agreed-upon value.
 Specificity measures the ability of a test to achieve a negative outcome for persons who do not have
the disorder for which it is being measured (also known as the "true negative" rate).
 Recall or Sensitivity is the ability of a test to detect patients with a disease correctly. Sensitivity
evaluates how much a test accurately produces a positive outcome (also known as the "true positive"
rate) for those who have the disease for which it is being tested.
 Precision is described as "the performance of being exact" and relates to the proximity of two or more
measurements, regardless of whether they are correct or not. Precision measurements have the
potential to be inaccurate.

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [88]
Artificial Intelligence in the 4th Industrial Revolution