Genetic Algorithm-Based Hyperparameter Optimization For Diabetes Type Prediction Using SVM And LSTM

Abstract

Diabetes is a chronic illness that needs the most accurate and efficient predictive algorithms for early detection and management. This paper utilizes Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) networks to predict diabetes from the Pima Indians Diabetes dataset. The model takes an enhancement through genetic algorithms that optimize hyperparameters using HPO. The SVM model was tested with four kernel functions, which included linear, polynomial, radial basis function (RBF), and sigmoid, while LSTM model training occurred with Adam, SGD, and RMSprop optimizers. The data preparation process began by handling missing value cases, applying feature scaling, and adjusting the unbalanced data classes after dividing the dataset for training 70% and testing 30%. The experimental outcomes validate the HPO method as a practical improvement for prediction accuracy. Both optimized SVM models using RBF kernel and LSTM model with Adam optimizer produced high accuracy rates, but the SVM model delivered 94%, and the LSTM model reached 90%. This study's findings highlight improved machine learning by optimization methods for enhancing diabetes prediction precision, which supports the development of trustworthy diagnostic healthcare instruments.