Generalized Denoising Coder with Multi-modal Approach for Handling Incomplete Information in Public Health Big Data Analytics

Abstract

A data-driven health model incorporates many missing and incomplete values, and the effectiveness of a health model that relies on characteristics not captured by the consumer is low. A Deep Learning (DL) health paradigm aims to improve the precision of learning weights. A DL-based health model is trained by optimizing the weights to maximize accuracy. If the learning process is excluded, the accuracy of applying a DL-based health model to the user's circumstance may be reduced, and it becomes the responsibility of people to manage Public Health Big Data Records (PHBDR). To handle incomplete information, this study suggested a generalized denoising coder with a multi-modal approach in public health big data records (GDC-MM-PHBDR). The suggested technique uses a Generalized Denoising Coder (GDC), which estimates the output values following the relevant input values using Neural Networks (NN). The National Health and Nutrition Examination Study (NHNES) provided data for this investigation. This approach allows for the estimation of incomplete information in the PHBDR. Multimodality, which the PHBDRs provide, enables compiling data for a single item from many sources. The layout of the GDC is multi-modal. The GDC-MM-PHBDR approach continuously demonstrates excellent performance, attaining maximum accuracy across all noise levels. It begins at 98.31% accuracy with a noise factor of 0.05 and significantly decreases to 94.5% with a noise factor of 0.25.