Enhancing Diabetes Prediction and Classification Using the Bidirectional Neighbor Graph Algorithm

The global prevalence of diabetes, a chronic health condition with diverse implications, necessitates improved prediction and classification methods. In this research, we propose a novel framework employing the bidirectional neighbor graph (BNG) algorithm to enhance diabetes prediction. By leveraging graph-based semi-supervised learning, we compare BNG with existing systems, thereby improving data structure modeling. The BNG algorithm addresses missing data and aims to optimize predictions for individuals with diabetes. This innovative approach holds promise for advancing diabetes research and creating more accurate prediction models for this condition. The methodology establishes a network connecting nodes to their nearest neighbors in both forward and backward directions. The evaluation of the model performance reveals an AUC (Area Under the Curve) score of approximately 0.86, demonstrating its efficacy in distinguishing true and false positive values across diverse classification thresholds. Moreover, BNG models effectively capture comprehensive and distinct features from the input data, resulting in improved classification performance. Additionally, the BNG method showcases computational efficiency, making it highly suitable for large-scale applications.