Predictive Modelling and Healthcare Monitoring System Using Enhanced Variational Auto Encoder Model

Obesity is a growing public health issue resulting from physical inactivity and poor nutritional
practices. Traditional health monitoring approaches are often subjective and inconsistent,
making early-stage obesity classification challenging. Accordingly, we propose a deep
learning (DL)-based framework using an Enhanced Variational Auto Encoder (EVAE) for
accurate multi-stage obesity detection. The framework consists of three main phases: dataset
collection, feature selection, and classification. InceptionNet (GoogLeNet) is adopted to
perform feature extraction and isolate key discriminative features within the dataset to enhance
input representation. The Enhanced Variational Auto Encoder (EVAE) is then used to learn
latent feature representations and classify obesity stages more effectively. To further optimize
model performance, the improved sand cat Swarm Optimization (ISCSO) algorithm is
integrated the fine-tuning in EVAE’s key hyperparameters. The proposed method significantly
outperforms existing models and achieves a high classification accuracy of 97.65%,
demonstrating its reliability and effectiveness in automated obesity stage prediction and
healthcare monitoring