Neural Bio Event Prediction With Time and Risk Outputs

The prediction output is segmented into multiple distinct components: a binary prediction indicating whether a bio event will occur within a predetermined time period, and a continuous prediction indicating the expected time of occurrence. This segmentation allows each component to serve specific clinical decision-making purposes while collectively providing comprehensive prediction information that resolves the information loss problem.

The prediction system transitions from a single-dimensional binary output to a multi-dimensional output structure that includes both temporal information (expected time of occurrence) and probabilistic information (likelihood within predetermined time). This dimensional expansion enables clinicians to prioritize patients based on both when and how likely a bio event is to occur, significantly improving ease of operation for patient triage.

The artificial neural network model is pre-trained to simultaneously generate both binary and continuous prediction outputs during the training phase. This preliminary action of learning multiple prediction tasks concurrently allows the model to produce comprehensive prediction information without requiring additional computational resources during actual inference, as the multi-output generation is already embedded in the model architecture.