UATT

About

An AI-based system for early sepsis prediction that augments patient risk assessment in clinical settings. This uses a Temporal Transformer model architecure to predict sepsis in real-time. The proposed framework incorporates masked self-attention to handle variable-length and irregular sequences, along with explicit missingness encoding to preserve informative absence patterns in clinical measurements. Predictive uncertainty estimation is integrated to improve reliability in high-risk clinical decision support. Presented at the 26th Philippine Computing Science Congress in Davao City.

Features

• Early sepsis prediction
• Real-time monitoring
• 71% accuracy