First, i test the quality of the sleep tracking against a scientific eeg, monitor second ill check, the heart rate accuracy and finally, i do a step counting test. As always, i do not want to waste your time so time stamps are in description below and also on the timeline. Music Applause, Music, hello, everyone for those of you that are new to the channel. My name is rob and im a postdoctoral scientist specializing in biological data analysis now before getting to the tests id like to provide the most important background. Information on the me watch light in under 30 seconds. The mi watch light has up to 9 days of battery life with typical usage as built in gps. Is water resistant up to 5 atmospheres and has 11 sport modes? It has a 1.4 inch square touchscreen display now the sleep tracking only tracks, night time, sleep and therefore daytime sleep, dosing and naps cannot be tracked. Importantly, the mii watch light only tracks deep sleep, light sleep and time spent awake, so it cannot track rem, sleep. The me watch light is relatively cheap and i bought it for about 50 euros. However, my channel is not so much about listing features. Instead, on my channel, i try to test the accuracy of different measurements. Now, over the last few months, the me watch light and its bigger brother, the me watch have been some of the most requested devices. I recently tested the bigger mii watch and today ill test.
The mi watch light in this video ill do a set of scientific tests and since im in the unique position to test the accuracy of the sleep tracking algorithm, i want to start off with this sleep test. For the sleep comparison, i wore the mi watch light to bed for three nights. At the same time, i also wore this portable scientific eeg device and i recorded myself using an infrared camera. The eeg device can actually measure your brainwaves and muscle movements and is therefore ideal for measuring your sleep stages. I manually went through the recording of the eeg and scored each part of the knight for the different sleep stages. Now i couldnt find a way to export the data from the mi watch light, so i ended up manually copying the data and loading it into the programming language i use now to the results i obtained lets. First, have a look at the accuracy over the three individual nights, after which ill do a statistical overview analysis. Here we see the first night i recorded on top. You see the sleep stages as they were recorded using the eeg device along the horizontal axis. We have the time of night and, as you can see, i went to bed a little after midnight on the vertical axis. You have the different sleep stages, that being deep sleep, light sleep, rem, sleep and awake. The sleep stages are plotted in the same order that are usually displayed in research on the bottom.
You can see a similar plot, but now for the sleep stages as they were recorded using the me watch light, as i mentioned in the intro. The me watch light does not track rem sleep, so i cannot check for this stage if we first look at deep sleep. According to the eeg, which i marked here in purple, we see that most of what was deep. Sleep was also marked as deep sleep by the me watch light. However, the mi watch light also detected a lot of extra deep sleep in the last two thirds of the night. That was not actually there looking at light sleep which are marked here in cyan. We see that some of the light sleep i had was indeed detected by the mi watch light. However, a lot of the light sleep was actually seen as deep sleep. Looking at rem, sleep, which are marked here in red. We see that what was rem sleep was mostly detected as light sleep by the me watch. Light awake times were detected pretty accurately by the mii watch, with it picking up on both of the awake moments, and these were also roughly of the right duration. Detecting the moment i fell asleep or slightly delayed, as you can see here in yellow brown, where i detected me falling asleep about 10 minutes later than i actually did. It was spot on. However, in detecting the moment i woke up. This is the next night i recorded again.
The deep sleep i had was mostly detected as being deep sleep, but a lot of extra deep sleep was detected as well, which means that a lot of the light sleep i had was actually detected as being deep sleep by the me watch. Light ram. Sleep was again mostly marked as light sleep by the mii watch light again, we see a slight shift in it, detecting me falling asleep where the mii watched light, detecting me falling asleep slightly too late, detecting my wake up time was again spot on now. This is the last knight i want to look at, and here we see that the deep sleep detection was a bit worse with the me watch, light missing a lot of the deep sleep at the beginning of the night and still detecting a lot of extra deep Sleep later, which again means a lot of light, sleep was marked as deep sleep by the me watch light and now what was actually ram sleep is either light sleep or deep sleep according to the mi watch light, and for this knight, detecting the sleep start was A lot better and also sleep and detection was really good. So far, the sleep tracking of the mi watch light does not really impress me. It does not appear to be completely random. However, it detects a lot of extra deep sleep, which was not really there to get an even more objective view of the results lets calculate some statistics regarding the consistency between the sleep stages of the mi watch light and the eeg device.
However, first a quick side note if youre interested in the latest updates on the wearables im testing consider subscribing to my instagram and my weekly newsletter. Of course, youd also make me really happy if you subscribe to this youtube channel now enough self promotion lets see what the overview statistics say. First lets look at the total percentage of each sleep stage, the eeg device and the mi watch light predicted. Here i display those percentages for the eeg device on the left and the mi watch light on the right again. The mi watch light does not predict rem, sleep, so theres no row for this overall. The mi watch light predicts about the correct amount of light sleep and also the weight. Detection is pretty good, however, theres a lot of extra deep sleep. Of course, all of this is slightly difficult to judge, because theres no rem, sleep more important than these total percentages is checking if the mi watch light predicts the correct sleep stages at the right time and thats. What i displayed here on top, we have the sleep stages according to the eeg device and on the left. The sleep stages according to the mi watch light now. Each column here sums the 100 percent, meaning that we can see what percentage of each of the actual sleep stages was recorded as each of the sleep stages by the me watch light. First, looking at deep sleep, we see that about 60 percent of what was deep.
Sleep was also predicted as deep sleep by the me watch light and all the rest of it was predicted as being light sleep for light sleep. We see that almost 70 percent of what was light. Sleep was correctly predicted as light sleep, and the remainder was predicted as deep sleep. Interestingly, more than 80 percent of what was rem, sleep was predicted as being light sleep by the mii watch light. This means that the light sleep according to me watch light fits better with rem, sleep patterns in real life than with light sleep patterns. Since rem, sleep matches 82 percent with light sleep. According to the me watch and light sleep matches 69 with light sleep according to the me watch. Finally, awake detection was pretty good with 80 percent of the awake time correctly predicted as awake time. The remainder of the awake time was predicted as light sleep, which makes sense, of course, since this is a lighter sleep stage. Finally, i want to check if the mi watch light detected me going to bed and waking up at the right time, which we already saw was a small problem based on the individual knights on the vertical axis. We have the dates of the nights i tested the miwatch light and on the horizontal axis, we have the time difference between the eeg device and the mi watch, light for waking up in yellow and falling asleep in blue. So a positive number means hes.
Detecting me as waking up or falling asleep later than in reality and a negative number means he detected me as waking up or falling asleep earlier. As you can see, the me watch mostly detects me waking up at the right time. However, as you can see, based on these blue dots here, it does tend to see me falling asleep about five or ten minutes later than i actually did for the next set of tests. Lets have a look at the heart rate, accuracy of the mi watch, light to test the heart rate accuracy. I will compare it against the polar h10 ecg chest strap, which is generally considered to be one of the most accurate consumer devices available for heart rate measurements. I wore both the mi watch lite and the polar h10 ecg chest strap for 4 spinning sessions, 2 bike rides and 2 weight lifting sessions. That way, i can check my heart rate at different heart rate. Ranges lets start off with the accuracy during spinning. Here i displayed an overview of the heart rate accuracy during spinning each dot. Here is a single heart rate measurement with long the horizontal axis, the value according to the polar h10 ecg chest strap and on the vertical axis, the value. According to the mi watch light, the blue line indicates perfect agreement, so any measurement along this line at roughly the same value for the polar h10 and the mi watch light. The red line indicates those measurements where the value according to the watch light, is half of the actual value according to the polar h10.
The reason i added this line is because, in the past ive seen that many devices measure half the actual heart rate when they make a mistake, the more measurements there are in a certain area, the darker black, the color. As you can see, the mi watch light performed pretty well during spinning, as most measurements are along the blue line. However, there are still some clusters of points away from the blue line, so lets have a look at the individual training sessions to see why this is here, you see the first spinning session along the horizontal axis. We have the time and my heart rate is along the vertical axis in blue. I plotted my heart rate. According to the polar h10 ecg chest, strap and in red is my heart rate. According to the mi watch light, as you can see, i took four short breaks in the spinning session, where my heart rate would dip for this first spinning session. The performance of the miiwatch light is pretty okay. It mostly overlaps with the measurements of the ecg chest. Strap but it does show some deviations. You can see that sometimes at the beginning of a segment, it needs some time to catch up with my increase in heart rate, this is visible to varying degrees in three out of the five segments. Looking at this second training session here, we see something similar with a mostly good overlap, but sometimes a delay in an increase in heart rate for this third spinning session, the overlap is already a lot better and finally, in this fourth spinning session, we again see some Issues with the delay in detecting an increase in my heart rate, especially here in the middle.
So far, the heart rate, accuracy is okay, but it does show some issues. Lets now take a look at cycling outside which are recorded while commuting to and from work. If i cycle outside, there are a lot more bumps and i also tend to sweat a bit more in the sun which might actually influence the accuracy of the mi watch. Light lets take a look now. This is an overview of those measurements, so a similar plot to before we saw for spinning, but now for cycling outside and though the deviation is a bit bigger. Its not bad. Most points are still along. The blue line lets now take a look at the individual commutes to see what these look like here we see my trip to work on my bike again in blue is the polar chest strap and in red is the mirrors light? As you can see, we see similar problems to the ones we had when spinning the agreement is mostly good, but sometimes theres a delay in detecting an increase in heart rate on the way back. This looks similar though it might even be slightly worse. Next lets see how the mi watch light perform during weight. Lifting now weightlifting is notoriously difficult for wristworm devices, because during weight lifting i flex the muscles and tendons near my wrist. This makes it hard for a device to actually detect the sudden changes. In my heart rate, lets take a look. This is an overview of my heart rate accuracy similar to before, but now for weight lifting.
Of course, the average heart rate is much lower during weight lifting than during cardio workouts. You can see here while there are a lot of points along the blue line, theres also quite a few points below the blue line, meaning the mi watch light detected at too low heart rate. This was especially true in the higher heart rate. Ranges lets check if we can find out why this is based on the individual training sessions. Here we see the first weightlifting session again in blue is my heart rate. According to the chest, strap and in red is my heart rate. According to the mii watch light, and you can see theres a pretty big disagreement between both devices, the mi watch is not able to pick up on the increase in heart rate that accompanies each set that i did and we can see the same for this next Training session right here where it was able to follow my overall patterns, but not the peaks and heart rate that come with each of the sets. I did overall my feelings on the heart rate. Accuracy of the mi watch light are a bit mixed. It had some issues during spinning where it had a delay in picking up an increase. In my heart rate, the me watch, which i also recently reviewed, was much better at this. However, during cycling outside the mi watch, light performed better than the mii watch and im still not sure why this is overall, the performance of the mi watch light was not terrible, though the mi watch light also features a step counter to see.
If discounts my steps accurately, i went out, and i took exactly 8 000 steps in segments of 1 000 steps to get an accurate step count. I manually counted my steps using this tele counter. Lets. Take a look at those results, as i mentioned, for the step. Counting test i went out, and i took 8 times exactly 1 000 steps. I alternated holding the tally counter in my left and right hand for each set of 1000 steps, which is what the right and left labels refer to now. These are the actual steps counted by the mi watch light and, as you can see, they were really close to the actual 1000 steps i took for each of the 8 segments. It was never more than 22 steps off now. Just to put it into perspective. Here are also the steps counted by the mi watch and the honor band 6. I wore at the same time for the first 4 segments, as you can see, they generally performed about equally well. Only the honor band 6 had a bit more deviations in the number of steps counted, but this is still very minor. Next i tested if the miwatch light ever gives any false positive steps. With that i mean, does it count any steps when its not supposed to count steps now to test this? I wore it while cycling to work and while exercising on my home trainer. That is what i will display here on the left.
I will show the results for cycling outside and on the right, the results for riding on my home trainer. Now here you can see the results from me cycling outside to work which takes me about 20 minutes. Each row is a single time i cycle to or from work and the left column is a total amount of steps counted and on the right. I converted this to the number of steps per minute that the mi watch light counted as you can see it counted. Almost no steps while cycling at most it counted about 3 steps per minute, which is much less than while walking, which would be around 100 steps per minute. And we can see something similar for while im exercising on my home trainer on the top right right here. Where it counted at most 3 steps per minute overall, the step counting of the mi watch light is really good. It counts the correct number of steps while walking and counts almost no steps while cycling or training on my home trainer. The only thing i still want to test is, if it counts any steps, while walking with the bike in my hand or while pushing a stroller ive heard from many of my subscribers, that some watches count no steps while pushing a stroller, and it might be that, If the watch is in a certain orientation, for instance, perfectly horizontal, it will just count zero steps, no matter what, overall, i would say that the me watch light is not bad, but also not great.
This sleep tracking was good for roughly tracking when you go to bed and when you wake up, and also for tracking your awake moments. The deep sleep tracking was not great, with the me watch, light tracking way too much deep sleep. The heart rate tracking was also not terrible, but it did often show a delay in tracking an increase in my heart rate. Finally, the step counting was good with the mi watch light. It tracks the correct number of steps while walking and it only counts a few false positive steps. So should you buy the me watch light? Well, it would not be my first recommendation, but for the price, its not bad, compared to the me watch, it misses a lot of features, feels much cheaper and also has worse sleep tracking. So if youre willing to spend a bit more, i would definitely recommend the me watch over the mi watch light. If you want the best sleep tracking from a watch, get a fitbit device like the fitbit charge, 4 fitbit sense, inspire 2 or lux the within sleep. Analyzer, which you put under your mattress, also performs reasonably well. The whoop strap has good heart rate tracking and also good sleep tracking. However, this is a subscription service, so much more expensive. Now, if you want a wrist, worn wearable that tracks your heart rate and spo2 accurately under all circumstances, then you should get an apple watch. Finally, i should mention some of the limitations of the data that i showed here.
First of all, i just tested the watch for a limited number of days and just on me and will be interesting to see how it performs on others. Also to do a full sleep comparison. It would be good to test the watch against the full scientific polysomnography setup. We actually assembled the polysomnography device using openvci components and were now working on getting this software functional. In my videos, i do scientific tests on different devices like the ordering, the fitbit and the scan watch and, in the end i hope to use tracking to improve my life. So if you like that, subject and like this video consider subscribing to my channel and also consider giving it a thumbs up, because it makes it easier for other people to find my videos. Thank you so much for watching.l6EhB_P6o5A
