Hyperparameter tuning of diabetes prediction using machine learning algorithm with pelican optimization algorithm

Abstract

Machine learning algorithms are employed in public health to forecast or diagnose chronic epidemiological illnesses like diabetes, which have global rates of transmission and infection. Machine learning technology may be applied to diagnostic, prognostic, and evaluation methods for a number of illnesses, including diabetes. This work presents a novel approach based on a novel metaheuristic optimization algorithm to improve diabetes categorization. 738 records were included in the final analysis of the main data, which was acquired in 2013 in accordance with the security protocols specified in the Declaration of Helsinki. This approach suggests a novel feature selection technique based on DBERDTO (Douche Optimization technique) and the dynamic Al-Biruni earth radius. A random forest classifier was used to categorize the chosen features, and the suggested DBERDTO was utilized to optimize the parameters. In this work, we investigate hyperparameter tuning for improved diabetes case prediction using the Pelican Optimization Algorithm (POA) in conjunction with the XGBoost machine learning technique. To prove the effectiveness and superiority of the suggested approach, it is tested against the most recent machine learning models and optimization techniques. The method's overall accuracy for classifying diabetes was 99.65%. These test results attest to the suggested method's superiority over alternative categorization and optimization techniques.