Automated Bias Detection within the Cardiovascular Disease Dataset using MapReduce Framework with Balance Measure

Abstract

Today, many fields rely on decision support systems, including health care for making appropriate decisions based on datasets. The decision support system, particularly in cardiovascular disease, is entirely dependent on the big data set, so if it is biased, it’s difficult to decide whether the person has a cardiovascular disease. Bias detection in cardiovascular disease datasets has become a complex task because of the direct processing of large data sets.  Another major drawback is that biases are detected on the set of attributes rather than protected attributes within the cardiovascular disease dataset which in turn increases computational cost as we know biases lie within protected attributes. Thus, it is a major challenge to identify the protected attribute from the set of attributes. Further, In the past bias identification was done manually using a statistical technique, which produced unreliable results i.e. minimum bias value related to cardiovascular disease. Considering all these challenges, we introduce a pioneering framework designed for automated bias detection within extensive cardiovascular disease datasets. Within our proposed methodology, we identified the protected attribute, namely gender, utilizing the capabilities of the MapReduce framework. Further, the balance measure approach has been used on the protected attribute of the cardiovascular disease dataset to detect the biases. The comparative results reveal that the detection of biases on protected attributes outperforms the existing works in terms of bias value, accuracy, precision, and F1 score which are 28%, 72%, 73%, and 81% respectively.  These metrics collectively indicate the superior performance of the proposed methodology.