Optimized Deep Learning Framework for Clinical Data Classification Using Firefly-Enhanced Stacked Sparse Autoencoders

Diabetes is a chronic metabolic disorder characterized by sustained high blood sugar levels, which frequently cause complications, including neuropathy and cardiovascular disease. Due to the complex and nonlinear nature of clinical data, accurate and timely prediction is challenging. Traditional approaches struggle to generalize or extract rich features from low-resolution datasets. In this paper, a hybrid deep learning model (FA-SSAE: Firefly Algorithm-based Stacked Sparse Autoencoder) is proposed to improve diabetes classification using the Pima Indians Diabetes dataset. Data is synthesized using Variational Autoencoder (VAE) developed data augmentation and deep features are extracted using SSAE. The model achieved 91.67% accuracy, 96.38% precision, and 98.75% recall; results that significantly outperformed several state-of-the-art methods. The results demonstrate the robustness and reliability of the proposed approach. Its lightweight architecture can be deployed in resource-limited environments, providing value for mobile or embedded systems used in remote clinics. This research advances the development of scalable and accessible tools for diagnostic detection of diabetes in the earliest possible stages to aid in unsupervised clinical care.