Can Wearable Tech Accurately Track Your Menstrual Cycle? New Study Finds

Emerging research shows machine learning could accurately predict menstrual phases using physiological signals captured from a simple wristband, no self-tracking required. But how reliable are the models, and what does this mean for the future of menstrual health?

Tracking your menstrual cycle is often laborious, requiring calendars, symptom diaries, basal body temperature logs, and ovulation test strips. But what if your wearable could do it for you, automatically?

A study published by NPJ Women’s Health in May 2025 suggests that menstrual phase identification might be possible using machine learning models trained on physiological signals, like skin temperature, heart rate, and electrodermal activity, collected passively from wrist-worn devices. With accuracy rates reaching as high as 87% in some models, this approach could offer a less intrusive, more personalised alternative to traditional tracking methods.

But the research also reveals where current models fall short, and where more work is still needed.

The Promise of Passive Tracking

The study analysed 65 ovulatory cycles from 18 participants wearing Empatica E4 and EmbracePlus wristbands for two to five months. These devices collected a range of data including:

• Heart rate (HR)

• Interbeat interval (IBI)

• Electrodermal activity (EDA)

• Skin temperature

• Accelerometry (ACC)

  
  

Importantly, participants were not asked to manually track or input symptoms beyond confirming ovulation with an LH test and logging menstruation days, dramatically reducing the burden of self-reporting.

Using data collected automatically from wearable devices, researchers trained machine learning models to identify different phases of the menstrual cycle:

• Period: The first phase when the body sheds the uterine lining, causing menstrual bleeding.

• Follicular phase: When follicles in the ovaries grow and the lining of the womb thickens, preparing for a possible pregnancy.

• Ovulation: The middle of the cycle, when a mature egg is released from an ovary.

• Luteal phase: The final phase after ovulation and before the next period, during which the body prepares either to support pregnancy or to begin menstruation again.

  
  

Two types of data processing techniques were tested: a fixed window (using set, non-overlapping time periods) and a rolling window (using overlapping, sliding segments that better mimic real-time application).

The chart shows how levels of oestrogen, progesterone, follicle-stimulating hormone (FSH), luteinising hormone (LH), and body temperature change over a typical 28-day menstrual cycle.

How Accurate Were the Models?

When classifying three phases, period, ovulation, and luteal, the random forest model delivered the strongest performance. Using the fixed window approach, it achieved:

• 87% accuracy

• 96% area under the curve (AUC) - a measure of how well the model distinguishes between phases.

  
  

This model performed consistently across different menstrual phases, with the ovulation phase predicted most accurately.

When the model was applied using a rolling window, more closely resembling how wearable data might be used in real-world tracking, the accuracy dropped to 68% when classifying all four phases. Still, even in this more challenging setting, the luteal phase was reliably predicted, and support vector machines (SVMs) showed the strongest AUC scores.

In more generalised testing (where the model was applied to an entirely new participant’s data), logistic regression performed relatively well but with lower accuracy, around 61–63% depending on the model and dataset.

Why These Results Matter

This study adds to a growing body of research demonstrating that physiological signals, such as heart rate variability, skin conductance, and temperature, change meaningfully across the menstrual cycle.

By capturing and analysing these changes, machine learning models could eventually provide real-time insights into where someone is in their cycle, without needing invasive testing or manual tracking. That could be game-changing for:

• People managing conditions like endometriosis or PCOS.

• Those tracking fertility.

• Anyone looking for deeper insight into how their cycle affects sleep, mood, or stress.

  
  

But It’s Not a Done Deal Yet

While the findings are promising, they also come with caveats. The sample size was small (just 18 ovulatory participants), and the physiological variation between individuals was considerable, making some phases, like the follicular phase, harder to classify accurately.

The study also didn’t account for demographic differences such as ethnicity or overall health status, which can influence physiological signals. And wrist-worn devices, while convenient, may suffer from inconsistent wear or interference from environmental factors.

Even the method of assigning phase labels, based on day-counting from menstruation and ovulation tests, might limit how well machine learning can truly “learn” the nuances of the menstrual cycle.

What’s Next?

To move from lab to wrist, future studies will need to:

• Include larger and more diverse participant groups.

• Track multiple cycles per person over longer periods.

• Explore individualised models that adapt to personal physiological patterns.

• Integrate advanced techniques like deep learning or transfer learning to boost performance.

  
  

But the direction is clear: wearable technology could soon offer a non-invasive, accurate, and personalised way to monitor the menstrual cycle, transforming not only fertility tracking but how we understand female physiology more broadly.

The Bottom Line

This study shows that physiological signals from the wrist hold real potential for menstrual phase tracking, especially when powered by smart machine learning models. With continued research, your wearable could one day be your most accurate cycle companion.