DRAGON-MINE: Deep Reinforcement Adaptive Gradient Optimization Network for Mining Rare Events in Healthcare

Key Points

• The aim is to address severe class imbalance in healthcare data by developing an effective mining technique for rare events.
• Developed a deep reinforcement adaptive gradient optimization network (DRAGON-MINE)
• Applied the model to healthcare data with severe class imbalance
• Analyzed the performance by comparing it to existing methods
• The DRAGON-MINE network effectively identifies rare events like sepsis and cardiac arrest
• Improved accuracy in detecting critical conditions compared to traditional methods
• Showed potential to reduce clinical consequences of overlooked adverse reactions