• Indoor Human Trajectory Identification with Mobile Sensors

I developed an algorithm from scratch to convert the messy accelerometer and gyroscope readings from mobile devices into understandable displacements and headings of the device holders, which makes it possible to identify the trajectory of a person according the floor map.

• Analyzing and Modeling NYC 311 non-emergency calls

I cleaned the NYC 311 data set (complex table with 52 attributes, 6M+ entries and tons of unexpected values), and put it into a MySQL Database, along with other external data sources, e.g. weather data of NYC. Then I did all the analyzing and modeling using Python connected to MySQL. The understanding from the analysis enables me to predict the future 311 calls at a high precision using the random forest.

• Modeling the Dynamic of Collaborative Human Behaviors

Building a universal model capturing two different dynamics in human online collaborative behaviors namely Initiation and Response based on temporal correlation. The result shows that about 60% of edits in Wikipedia have causal relationship with earlier edits.