Improved Diabetes Prediction with Reduced Feature Sets: Evaluating Feature Selection Techniques in Machine Learning

Diabetes is a chronic metabolic disorder characterized by elevated blood glucose levels, which can lead to various health complications. Manual assessments for diabetes prediction are often subjective, underscoring the need for automated systems. Machine learning (ML) techniques have been widely used for diabetes prediction. However, many studies rely on many features, resulting in increased complexity. In order to address this issue, this study compares various feature selection (FS) techniques applied to six ML classifiers, aspiring to examine their impact on diabetes prediction. The results demonstrate that the Random Forest (RF) classifier achieves the highest accuracy of 92.2% by utilizing only four features. By underlining the significance of employing a reduced feature set, our findings enhance the efficiency and effectiveness of diabetes prediction models. The implications of our research lie in the development of automated systems that accurately predict diabetes while minimizing the complexity and computational burden associated with an extensive feature set. Moreover, our work serves as a valuable reference for researchers and practitioners in the field of diabetes prediction, facilitating the development of streamlined and practical ML-based approaches.