Exploring Factors Influencing Intradialytic Hypotension and Hypertension in Small-Instances Hemodialysis Data: A Data Profiling Approach

The detection of Intradialytic hypotension (IDH) and hypertension (IDHTN) during hemodialysis treatment is necessary as these conditions can severely impact a patient's health and increase the risk of mortality. Early detection and intervention are crucial to prevent harm and ensure successful treatment outcomes. However, gathering hemodialysis data can be challenging and time-consuming, which may lead to a small dataset. Moreover, this problem will be more complex when it involves other issues, such as inconsistent and missing data. Thus, this study aimed to adapt the data profiling technique to solve the mentioned problems. The process involved three phases: structure discovery, content discovery, and relationship discovery. The data profiling technique created five datasets through weight vector categorization: (1) Full Dataset, (2) Mid Moderate and Low Moderate (MMLM), (3) Strong and High Moderate (SHM), (4) combination of MLMM and SHM and (5) Strong Only. Four classifiers, Naive Bayes (NB), Support Vector Machine (SVM), Linear Regression (LR), and K-Nearest Neighbor (KNN), were used to predict the occurrence of IDH and IDHTN, and their accuracy was observed and evaluated. The study found that KNN gave the best performance, with an average accuracy of 81.54% for all datasets, compared to the other three classifiers. The study demonstrates the effectiveness of data profiling techniques in improving the accuracy of machine learning models, even with small datasets. Future research could expand the study by including more factors and data to provide a comprehensive view of patients and predict the occurrence of IDH and IDHTN.