Abstract
Blood pressure measurement and prediction is an important condition for heart patients and people with heart problems and should be kept under constant control. In this study, based on the oscillometric waveform obtained from individuals using a cuff, the oscillometric waveforms are divided into three periods. These periods are; the first period from the starting point to the systolic blood pressure (SBP), the second period is between systolic blood pressure (SBP) and diastolic blood pressure (DBP), and the third period is between diastolic blood pressure (DBP) and end of the waveform. In the dataset used, the attributes obtained from the oscillometric wave envelope were subtracted for each pulse. On the dataset, the attributes of the beat corresponding to the systolic pressure point are labeled 1, and the attributes of the beat corresponding to the diastolic pressure point are labeled with 2. Other beats are labeled with 0. In the study, the dataset was first re-labeled. Systolic beats were labeled with 1, beats before systolic point, 2 with systolic, diastolic point including diastolic point, and 3 with a diastolic point. After re-labeling, 350 measurements, 300 measurements were divided into training and 50 measurements were divided into test data subset. Classifiers were trained with 300 subsets and the classifier model was produced. With the generated model, the classification of the pulse sequences in the test data subsets was performed. In the found label series, the first 1 to 2 label was marked as the systolic pressure point and the last 2 to 3 labels as the diastolic pressure point and the corresponding cuff pressures were estimated as systolic and diastolic pressure values. By classifying these periods, the systolic blood pressure (SBP) and diastolic blood pressure (DBP) values have been estimated using three classifier algorithms including k-nearest neighbor (kNN), weighted k-nearest neighbor (WkNN), and Bagged Trees algorithms. To evaluate the performance of the prediction algorithms, four different performance metrics comprising MAE (mean absolute error), MSE (mean square error), RMSE (root mean square error), and R2 have been used. For the estimation of SBP values using the kNN algorithm, weighted kNN, and Bagged Trees, the obtained MAEs are 3.590, 3.520, and 4.499, respectively. As for the estimation of DBP values using kNN algorithm, weighted kNN and, Bagged Trees, the obtained MAEs are 11.077, 11.032, and 13.069, respectively. The obtained results demonstrated that the proposed method could be used in the blood pressure estimation as the new approach.