<scp>MA</scp> ‐ <scp>AXGB</scp> : A Secure and Efficient Attention‐Enabled Auto‐Encoder‐Based <scp>XGBoost</scp> Ensemble Model for <scp>UTI</scp> Prediction in <scp>IoT</scp> ‐Fog Healthcare Network

ABSTRACT Urinary tract infections (UTIs) are among the most common health issues, severely affecting the kidneys and other functional organs, necessitating early prediction. With the rise in IoT‐Fog computing, healthcare data can now be collected and processed easily. However, earlier research possesses challenges in secured access and storage of data, ineffective selection of features, and improper predictions. Therefore, to address these issues, this research proposes a Multi‐head Attention‐enabled Auto‐encoder‐based XGBoost (MA‐AXGB) model for UTI prediction in IoT‐Fog Environments. The MA‐AXGB methodology integrates an Autoencoder architecture to refine and compress the health features, ensuring reconstruction and retention of the most relevant features. The multi‐head attention mechanism offers a considerable benefit by analyzing the most important elements of the UTIs, and the XGBoost classifier improves predictive accuracy and mitigates data overfitting through the strategy of decision trees. Moreover, the biometric authentication for secured data handling enables protected storage and authorized access within the IoT‐Fog environments. The performance of the MA‐AXGB model is tested on the two benchmark datasets, gaining 97.75% accuracy, 98.52% sensitivity, and 96.31% specificity on the Urinary Tract Infection in Tanzania dataset. Similarly, the proposed method obtains 96.96% accuracy, 97.79% sensitivity, and 95.40% specificity on the Urinary Tract Infection in the Emergency Department dataset.