Future Work
Longer deployment	I am planning a longer deployment of the system to office workers. A longer deployment in an individual's office means more data, thus the training of the machine learning algorithms can be tailored to the user. Another benefit of a longer deployment is potential improvement of the productivity data. Subjective ratings is subject to users' moods; a longer study may reduce this variability. This study is difficult to deploy for a week, because it demands a lot from the user (users are required to document productivity every 15 minutes). I am currently looking for methods to reduce the load on the user.
Feature selection	For this study, I used all the features to train the classifiers. It might be the case that one or two of the features are sufficient to detect activity. It is also possible that other features from other sensors are better detectors of activity and this needs to be explored. A technique to explore which sensors may be good predictors is the Wizard of Oz method that Hudson et al. (2003) used to explore the feasibility of detecting interruptibility.
Temporal model	Activities are temporal and have patterns that can be exploited to improve detection accuracy. Many research projects (Oliver and Horvitz, 2002; Niu et al. 2004) have shown that human behavior can be modeled well using HMMs. Currently, the system uses non-temporal learning algorithms, such as Naive Bayes and Logistic Regression. By using a temporal model, the system can take advantage of the temporal properties of the activities.
Other applications	Monitoring completion of goals. Setting goals is a crucial element that can help people make changes with managing their time. This system can help by monitoring people's activities and comparing it to the person's goals. The system can then remind the person of their goals or to help with assessment of completion of goals. Visualizations of activity/productivity. Similar to the work of Begole et al. (2003) on rhythm awareness, we can explore what kinds of visualizations would work well in informing people of their daily activities and productivity levels. A good visualization should help the user make sense of the vast amount of data that is being recorded about them. Also, it should make help them prioritize activities throughout the day, weeks, and months.