Biosignals reveal where and how you concentrate best
Poor concentration hampers learning and work performance. Our current environment is often full of stimuli that distract us. They increase dopamine production in our brain, which makes long-term concentration difficult. Professor Caj Södergård and Senior Scientist Timo Laakko at VTT developed a machine learning model for the evaluation of concentration. In the next few years, smart devices may also be able to measure concentration similarly to how they now measure steps, pulse, sleep quality and stress.
Image: Christin Hume, Unsplash
For many, concentration problems are common, and the reasons for these problems are individual. They may be difficult to define on your own. When and in what situation do you concentrate best? What kind of teaching helps students concentrate? Only by measuring the level of concentration can the student receive concrete information on what kind of environment, situations and times are best.
Concentration is a feature that can complement the information provided by smart devices in the future. Eventually, a machine learning model can be used in smart watches, activity wristbands, smart rings or other wearable smart devices—why not in earrings. This feature could also be used by educational institutions and staff training companies to help create optimal, efficient and tailor-made learning materials and situations for students.
The development of the machine learning model was based on the ADECO project funded by Business Finland a few years ago, which examined what information biosignals indicate about learning situations and concentration. Researchers built a data collection system in a cloud service to predict the concentration of test subjects from their biosignals. The daily studies of sixteen students from Haaga-Helia University of Applied Sciences were monitored for one to two weeks. Data was collected from the students as a basis for the machine learning models. The researchers collected information about the students' biosignals using a smart wristband: body movements, skin conductivity, heart rate, heart rate variation and skin temperature. At the same time, the students assessed their own ability to concentrate with their mobile phones.
The researchers developed machine learning models based on both boosted regression tree and convolutional neural networks. The users' assessments of their own concentration levels were rather limited, but there was plenty of biosignal data. In order to utilise biosignals from the measurement periods during which the test subject had not assessed their concentration, a semi-supervised machine learning method was developed, which also significantly improved the results. The best model predicted the concentration level with only 1.7 percent NMAE error.
The models showed that skin temperature is the most accurate predictor of concentration. The skin's electrical conductivity and heart rate variation were also important measurement signals. The signals were the opposite of what would be seen in stress reactions. In fact, concentration increased as the skin temperature increased. Similarly, high skin conductivity, in other words, increased sweat, indicated a lack of concentration.
The results show that good concentration with studying requires long-term relaxation—often called flow. High concentration is not characterized by intensive efforts, which are typical of stress situations that have been studied extensively.
“We used biosignals in our study—pulse, skin temperature, electrical conductivity and body acceleration—the same signals gathered and used by the latest smartwatches and wristbands. Therefore, the application of our research results is a matter of software updates,” says Caj Södergård.
Reference: “Inferring students’ concentration levels in daily life using biosignal data from wearables“, https://doi.org/10.1109/ACCESS.2023.3260061