Abstract
During the first half of this project, we learned about Intel’s telemetry framework. The framework allows remote data collection from devices with Windows operating systems. Two important components of the telemetry framework are the Input Library (IL) and Analyzer Task Library (ATL). The IL exposes metrics from a device and the ATL generates on-device statistics from the data collected by the IL. In the second half of the project, we used pre-collected data provided by Intel that used their telemetry framework to create a classification model. Our goal with the model was to predict the persona of a user using their computer’s specifications, CPU utilization, CPU temperature, and time spent on certain types of applications. User personas were provided by Intel which classified if users were casual web users, gamers, communication, etc.. The classifications of these personas were done by Intel based on the amount of time users spent on certain applications based on their usage of different types of .exe files. For example, if a majority of a device’s time is spent on an application like Skype, they are most likely classified as a communication user. Similarly, if a user spends a majority of their time on the League of Legends .exe file, they are most likely classified as a gamer. After training multiple classification models, we were able to predict user personas with 64% accuracy using a gradient boosting classifier.
