1. I’ll capture a host-based data set on my own machine that represents user activities. That is, captured data by means of a sensor is going to be the base of normal behavior of legal user. Then, I’ll send those data to the coder.
2. Initial task of the coder is to analyze the captured data to design a profile to model my own behavior so that the captured data and the resultant model can be used to determine how often I conform to my own behavior. Any tools can be used.
3. Experiment by running the model against "test data" I capture on my machine and measure the accuracy of your model. I will, for example, run a data capture over 1 week, and choose some percentage of the data for training, say 80% of the initial portion of the data; test on the remaining 20%.
4. Plot the tests to determine the FP rate of sensor/detector. As part of test, I’ll have another person use my machine for his own work to determine if he behaves differently than me. I’ll let the coder know the start time and end time when this other user used my machine for the report part.
5. Please prepare a report, along with the design of system, and the performance results including ROC curves of system.
Sample data files are attached. It's an amateur and relatively elastic project, so coders can discuss me details.
Due date is [url removed, login to view] by midnight.