1.8 billion people in the world regularly menstruate (Rohatgi and Dash, 2023), and are not able to track their periods with conventional methods such as calendars or period apps. Oftentimes, conventional period tracking methods are not well suited to those with disabilities because they do not apply to those with irregular periods, are not accurate enough, and/or are difficult to use (Attia et al., 2023). Data safety is another concern for people who want to track their cycles, especially with the overturning of Roe v. Wade (Basu, 2022).
Our project aims to develop a bracelet that measures three vital signs (heart rate, skin temperature, and skin conductivity) and utilizes a machine-learning model to predict days until the next period. Our goal is to make menstrual health monitoring more accessible, safe, and reliable for all.
The current design aims to measure three vital signs (heart rate, skin temperature, and skin conductivity) and use that to predict the start of the next period cycle. The three sensors are embedded in a silicon bracelet because silicon is a type of insulator and will not interfere with the sensor readings. The vital signs are recorded by an Arduino UNO that stores six variables (average and current value of each vital sign). These six variables are sent to the Raspberry Pi via bidirectional communication. A machine-learning model was designed using Python. Decision tree algorithms were used to map the user’s vital signs to the phases of the menstrual cycle. The user is also provided with a button that they can press when they start and end their period, so the machine learning model can adapt to their cycle patterns. The predicted period start date is displayed on the LED screen. In the future, we will work towards reducing the size of the product and decluttering the wires to make it easier to use. Additionally, the device must be programmed to run without the need for a connection to a computer. We can take multiple steps to accomplish this. First, we can set up the Raspberry Pi to run headlessly so that the device doesn’t require a monitor. Then, we can trade the microcontrollers for an FPCB to significantly reduce the size. An FPCB would work best for a bracelet and would almost definitely fit a wrist better than the current model because, aside from there being no breadboard, the sensors would be smaller and built into the bracelet. Last but not least, the machine learning model can be made more robust and accurate by using advanced Lasso Regularization techniques and incorporating an optional feature that allows the user to input information about their menstrual cycle history.