Sarah - Summer Mentorship & Progress

Welcome back to my blog!

Time has flown so fast and it’s almost the 3M Final Event! I have been continuously creating and testing my Spark Care+ Prototype over the summer with the support and help of my mentor, Dr. Ann Fornof!!!

I have had a tremendous progress on building my prototype during the 3M summer mentorship. While I was testing the effects of music to human emotional changes through the Galvanic Skin Response and Photoplethysmography sensor, I needed to find a more efficient way to test the healing powers of music, the effect of different genres, and the effect of tempos.

I used a Placket Burman Design of Experiments to effectively see how genres and tempos generally affect the participant’s heart rate variability and the mean of GSR measurements. I hope to continue to test my prototype on as many people in the following weeks.

Creating my physical prototype was very exciting! Here are some pictures I took along the way.

Measuring my responses with the GSR and PPG sensors using Arduino. I used Arduino to communicate the sensor data to my Python machine learning program.



The 3M product for the wristband is a Hook and Loop style product that is versatile but sturdy and works well as a comfortable wristband.

Even while including standard error, music, has a visible emotional change! Additionally, the various tempos and genres (classical, pop, and rock music) I tested, had different effects to the participant’s GSR/HR measurements. The below is a graph from one of the participants for my DOE experiment.

One of the main aspects I have been focusing on lately is the machine learning code using Python while Arduino communicates the sensor results to Python. It was a challenging aspect of the prototype but felt so great when it finally worked. The regression that was most applicable to my data sets from the Placket Burman DOE was Ordinary Least Squares Regression. OLS regression finds the relationship between the GSR/HR sensor measurements and the music. In Ordinary Least Squares, we try to minimize the residual squared. Although music is not a numeric value, it can be interpreted in a way such that it is placed on the coordinate plane based on GSR/HR measurements, tempo, and genre by using matrices.

Using Python, I can find the coefficients form the OLS regression. With the machine learning code, Python can provide key information from the csv file containing music, genre, tempo, and the measurement results of the GSR and HR sensors. It indicates coefficients, standard error, R squared, and the skew.

Throughout the entire process, my 3M mentor has guided me and encouraged me to explore new approaches, applications, and 3M products. Dr. Ann also introduced me to other experts in the machine learning and music therapy fields who shared invaluable information for applications of their specialties. Dr. Ann and I met in weekly meetings and it was the highlight of my summer! I am so thankful for this opportunity of mentorship!

See you next time!