<![CDATA[This research addresses the challenge of feature selection in high dimensional medical datasets, where unnecessary or duplicated information can hide patterns and negatively impact model performance. The aim is to develop an efficient feature selection strategy using Fine-tuning Fusion Graph Convolutional Networks (GCNs) to enhance model accuracy and interpretability. The objectives include improving the medical data selection process, increasing generalization, and assisting healthcare professionals in making educated clinical decisions based on the most relevant factors. The study employs Joint Fine-Tuning Fusion Graph Convolutional Networks (GCNs) for feature selection in medical datasets. This approach entails creating several graphs to illustrate feature interrelations, amalgamating them into a cohesive representation, and optimizing the model to emphasize pertinent aspects. The L2-norm of the final embeddings dictates feature significance, directing the choice of the most critical features for enhanced predictive accuracy. The study's findings indicate that GCN-based feature selection improves classification accuracy, especially for the PIDD dataset, enhancing accuracy, precision, recall, and F1-score from 0.74 to 0.75. The Kidney Failure dataset exhibited near-perfect accuracy (0.99) prior to selection, whereas the heart disease dataset had a minor reduction in performance (from 0.81 to 0.80), highlighting the dataset-specific effects of feature selection. GCN-based feature selection improved classification performance, increasing the PIDD dataset's accuracy from 0.74 to 0.75, with no significant effect on the Kidney Failure dataset. Nonetheless, it somewhat diminished performance for the heart disease dataset. Subsequent study ought to enhance feature selection techniques by integrating dataset-specific optimizations and domain expertise to augment model precision and overall generalizability.]]>
Copyrights © 2025
