Performance Analysis of the Ada-Boost Algorithm For Classification of Hypertension Risk With Clinical Imbalanced Dataset

Abstract

High blood pressure, another name for hypertension, is a condition in which the pressure inside the blood arteries keeps rising. One of the most exciting fields of study is data mining, which is being developed to increase its acceptance in organizations, particularly in the health sector [5]. Nowadays, the AdaBoost method is a crucial feature classification approach in machine learning. AdaBoost assigns weights to the samples during the training phase that increase as the error rate rises and reduce as the error rate decreases. The classification process on the hypertension risk using clinical Healthy Family Index dataset, obtained of 8 attributes. Requires two partitions, 70% training data and 30% testing data. Binary classifications are applied “Normal” and “Hypertension”. The most important features of the 10 attributes are chosen based on the ranking value of the ReliefF feature selection algorithm, with the 5 best-ranked attributes. ReliefF feature selection can distinguish between classes and identify the most relevant feature with a value exceeding a specified threshold. The high confusion matrix results also show that the AdaBoost algorithm applied in this study has good performance. This is also supported by pre-processing data which is carefully processed in order to not affect the final classification result. The initial preprocessing step is to data cleaning, missing value handling, data normalization, feature selection using relief-F, imbalanced dataset handling using oversampling technique, classification process with adaBoost algorithm, test and score and confusion matrix.