Milestones
Milestone 1: Initial meeting with Drs Lattie and Saeb about Purple Robot and the CS120 dataset
Milestone 2: CITI certification
Milestone 3: Created project plan
Milestone 4: Received and gained access to the CS120 dataset
Milestone 5: Brainstorm of appropriate algorithms
Milestone 6: Visualization of raw data to understand potential relationships
Milestone 7: Understanding Sohrab's Matlab analysis
Milestone 8: Initial calculations and code for feature extraction
Milestone 9: Extraction of all features
Milestone 10: Analysis of different classifiers
Milestone 11: Improving performance through multiple means including removing partial workday, creating personalized models, and clustering.
Milestone 12: Analysis of results
Milestone 13: Paper and presentation
Future Steps
Gather more sensor data especially from a more diverse set of participants
While the 207 participants from the CS120 dataset led to a significant amount of data, we believe that collecting even more sensor data could improve the performance and generalization of our models. Removing data entries with missing feature values significantly improved model accuracy. Since removing these entries significantly decreased the dataset size, more sensor data could overcome our diversity and missing data limitations and confirm our exploratory results.
Make additional predictions from sensor data that may have a strong link to depression
There are additional predictions related to depression that could be explored within the CS120 dataset such as affect (stress, mood, energy, focus). Our study showed the promise of using mobile sensor data to make predictions and believe that further data analysis could result in other useful findings in helping to detect and treat depression and other mental health disorders.