Optimization Algorithms in Diabetes Prediction: A Comparative Work

Accurately predicting chronic conditions such as diabetes is essential for prompt treatment and improved patient outcomes. Using the UCI diabetes dataset, this work assesses four optimization algorithms: Adam, RMSprop, Nadam, and Stochastic Gradient Descent (SGD). The objective is to evaluate their efficacy in creating precise prediction models by empirical means. These algorithms have a significant impact on the stability and speed of convergence of deep learning models during training. With patient data, machine learning and deep learning models can be used to assess the risk of diabetes. Diabetes is a complex metabolic disease that requires early detection to prevent complications; however, the choice of optimizer has a significant impact on the models' efficacy. As the optimizer with the highest accuracy, precision, and F1 Score, Nadam stands out. SGD does exceptionally well in recall, while Adam and RMSprop do well as well. The choice of optimizer should take trade-offs between measures into account and be in line with specific therapeutic goals. They can be more effectively tuned as hyperparameters in diabetes prediction models. This work improves patient outcomes in the management of diabetes and healthcare risk assessment.