Smart devices held by billions of people all over the world contain hardware that is capable of mass data collection such as acceleration sensors, cameras, and fingerprint sensors. In this project, the problem of human activity prediction is outlined and analyzed. Using only simple features extracted from the time series data of smartphone/smartwatch accelerometers and gyroscopes, a classification model was made to identify what the subject was physically doing, such as walking, typing, eating, etc. Then, the variable that matters most when attempting to create a predictive model is identified. This project is noteworthy because it has the potential to save thousands of lives in the medical field or help fitness instructors monitor the progress of their clientele.
The "main.R" script containts the majority of code used in this project. The "tables.R" script contains the code used to make tables. The "graph_maker.R" script contains the code that creates graph of acceleration reading over time for each user, each activity, and each device.
The data from this project comes from the WISDM smartphone and smartwatch activity and biometrics dataset found in the UCI Machine Learing Repository.
A quick note: This was my first ever major project involving programming. This project was pursued during my studies as a mechanical engineering undergradute, and thus at the time I did not know coding conventions and best practices. This is something I am improving over time via application and other projects. If there are any questions about the code, feel free to ask!