A Weighted Membership Approach to Fuzzy C-Means for Healthcare Data Imputation

Abstract

Missing data is the absence of value from its place in any given dataset. It occurs for various reasons, such as equipment malfunction, erroneous entry, etc. Handling missing data in healthcare datasets is essential for proper analysis and decision-making. Practical healthcare actions is hampered by incomplete records, undermining the validity of ndings. To solve this problem, ensuring data integrity and creating thorough knowledge and robust techniques are necessary. By e ectively handling it, healthcare workers will improve the quality of analysis, which will eventually result in better decision-making and increase the overall e cacy of healthcare systems and treatments by implementing e cient strategies. The paper proposes an algorithm based on Fuzzy CMeans (FCM) with a weighted membership approach that outperforms the available techniques. The contributions include a novel methodology for estimating missing values in healthcare datasets, retaining the dataset's underlying distribution while maintaining vital information, proposed work ow, and handling numerical and categorical data types. This multi-step procedure yielded more accurate results than existing methods: Mean imputation and Fuzzy C-means with Genetic Algorithm (FCMGA). The missingness mechanism considered is missing at random (MAR) and missing not at random (MNAR). The experimentation is carried out on two benchmark datasets to assess the e cacy of the proposed approach. The proposed method improved MSE and NRMSE scores on Parkinson's and Heart disease datasets.