UTFacultiesEEMCSDisciplines & departmentsPSEducationAssignment: Making activity detection more accurate

Assignment: Making activity detection more accurate

Making activity detection more accurate



With increase in population of elderly people, the advancement in monitoring techniques is also necessary. For that, the use of channel state information (CSI) for monitoring the daily-life appears to be a potential solution. For effective implementation of these systems in real-life, the robustness of the algorithm must be ensured.

Problem statement

Wi-Fi-CSI has potential to detect various human activities. A dataset having various hand and leg activities was built to test possibilities of CSI in small scale activities recognitions such as tapping, kicking etc. To analyze the performance of this dataset multiple machine learning models needs to be trained and tested.


Here you will first understand the rationale behind multiple machine learning algorithms. Then to enhance the performance of the activity classification, different algorithms (CNN, LSTM, bi-LSTM etc.) with hyperparameter tuning needs to be trained and tested.


40% Theory, 40% Implementation, 20%Writing.


Nikita Sharma (n.sharma@utwente.nl)