Work Breakdown Structure Overview

Detection Algorithm Logic low

Reporting Algorithm Logic Flow

Recording Algorithm Logic Flow

Machine Learning Logic Flow

Each training session will use:

Min, Average, & Max X,Y,Z acceleration over last 5 mins, 1 min, & 10 sec

Must remove gravity sdk has helper functions

Then clamp value in m/s2 between -1 & 1, add 1 & divide by 2 to get value between 0 & 1

Min, Average, & Max X,Y,Z gyroscope rotations over last 5 mins, 1 min, & 10 sec

Add 360 to given degrees then divide by 720 to get value between 0 & 1

Min, Average, & Max 10-second heart rate measured over last 5 mins, 1 min, & 10 sec

Clamp 30-250 beats/min; subtract 30 & divide by 220 to get value between 0 & 1

Total of 3*(3*3*3) + 3*(3*3*3) + 3 = 57 input nodes
Single output node is 0 if no seizure during time period, 1 if seizure during time period.

Machine Learning Algorithm: The Objective

Are you curious about what machine learning does? Check out the image below.
This picture portrays two sets containing an N-dimensional element and a new N-dimensional element.

Which set does the new element belong to? This is what machine learning does for us behind the scenes.

Machine Learning Algorithm: Training

A small portion of the total network: x-axis rotation and its small role in detecting a seizure.
Training is computationally intensive, and occurs on the external server, using the Gradient Descent training algorithm.

Machine Learning Algorithm: Evaluation

Evaluation is computationally trivial (adding/multiplying ~200 numbers)
It occurs on the smartwatch in real-time as 10-second bathes of data that are read from sensors.