Brainstorming

Accelerometer-based projects

From what I have seen in previous projects, accelerometer-based biometrics and local user activity (walking, climbing stairs, sitting down, etc.) recognition have been covered. While Ramachandran (from 2013) does not present a good study of user identification, Molay, Kouang, and Tam do so in "Learning Characteristics of Smartphone Users from Accelerometer and Gyroscope Data" (same year). They are able to identify individual users with relatively high accuracy:

Furthermore, 63 of the 75 individuals were correctly identified 100% of the time, 10 of the 75 individuals were identified 80% of the time and the remaining 2 were correctly identified 60% of the time. When identifying user gait data from either the fixed or mounted position, we correctly identified the user associated with the gait data 76% of the time. Our accuracy of 96% is on waist-mounted gait data is on par with the work achieved by prior researchers [4,12], which is impressive considering our data set contained significantly more individuals. However, we also note there was a recent Kaggle competition [5] that showed accelerometer data could be used effectively as a biometric.

Given this, maybe the better direction would be to analyze longer-period activities, specifically modes of transportation. This has not been done in previous years and, while the dataset would not be as large as the kaggle set, we could probably assemble one of sufficient size to do an analysis using the sensor kinetics app. What do you two think? I am willing to proceed with this idea and am of course open to others.