This research topic can only be taken by Thesis route and Project route student.
This research topic include system and mechanism development. The project outcome needs to be examined by well-designed experiment for quanititative and qualitative data analysis (for Thesis route student) and pilot testing for performance evaluation and comparison (for both Thesis and Project students).
Some elderly people may stay in the house all day long. This topic is trying to find out a mechanism to analyze the activities of the elderly people's daily life data which is collected by some ADL sensors. Furthermore, we can build an alarm system based on the ADL analysis results. The alarm system can use association rules, case-based reasoning, and/or pattern comparison, to trigger the notification processin order to prevent anything happened to the elderly people or give elderly people medical suggestions in advance.
There are several kinds of ADL research, first of all, what is ADL? As you know, when we are at home, we will move around in bed rooms, working room, kitchen, rest rooms, living room, and basement. Besides, we will do a lot of things by touching, opening, closing and moving things, for examples, cups, ovens, and closets. So, do our behaviors have patterns? For example, if today is working day, friday, or weekend, maybe, perhaps, our behaviors are pattern-able.
So, if our behaviors do have patterns, can we connect it to the health care? For example, if today is working day but your behaviors are analyzed and mapped to the 'catch a cold' pattern, then you might catch a cold! It is easy thought, right? Also, think about it, if a person doesn't have behaviors of getting up in the midnight, however, recently, he/she gets up and goes to the rest rooms in the midnight quite often, what's happened to him/her? It may be a health warning that his/her body wants to tell him/her.
More specific, if we apply this kind of ADL analysis research to elderly people who live independently or people who have long-term care requirement, can we reduce the work-loads of caregivers and make the caregivers have more time and ability to take care more people? ADL may be a solution of the shortage of medical resources and healthcare professonals.
Before we can analyze the ADL data, we need to have ways of collecting the data. At present, there are several ways to collect people's ADL data:
1. camera (include Kinect for Windows)
2. wearable sensors: may be some accelerated motion detectors or RFID senders/receivers
3. small number of sensors: usually include infra-red motion detectors and accelerated motion detectors
4. dense sensing approach: usually use hundreds sensors, including RFID tags, wireless motion sensors, wired sensors (for detecting switch's on/off, humidity, pressure, temperature, light, gas, current, water flow, and so on.)
After we have the data, as we know, the data should be analyzed from the sequential viewpoint. There are some researches can analyze data from the sequential viewpoint, for examples, different Hidden Markov Models (HMMs), Multiple Behavioural Hidden Markov Model (MBHMM), Sequential Data Mining, Time-Series Data Mining, Sequential Mobile Access Pattern (SMAP), and so on. A student of mine has done a comprehensive survey on these approaches and you may do further survey based on his essay (http://dtpr.lib.athabascau.ca/action/viewdtrdesc.php?cpk=495&id=48863), because a research work needs to have in-depth and comprehensive survey on at least 40 to 60 literature (and one third should be within three years and half should be within five years) to support the research's motivation, method, evaluation, findings, discussion, and conclusion.
Dr. Stephen S. Intille in MIT has real ADL data which you may ask if Dr. Intille could share it with you for testing your proposed algorithms and mechanism. The ADL data consists of 22 days of sensor readings from August 22, 2006 to September 18, 2006, has 784 sensors in a 1000 square foot lab, the PlaceLab, consists of a living room, dining area, kitchen, small office, bedroom, full bath and half bath. There are 126,747,912 sensor readings.
Another student of mine has designed and proposed a method of analyzing and extracting behavior patterns and formulating and representing the activities and sub-activities. You can find his work at
http://dtpr.lib.athabascau.ca/action/viewdtrdesc.php?cpk=0&id=48850
The method we discussed and proposed in Mr. Lam's research so far includes:
a. the relations among behavior patterns, low-level activities, and high-level activities
For examples, a behavior pattern may be considered as 'doing snack' if it happens in midnight, but it can be considered as 'prepare lunch' if it happens at 11:30 AM, right? Also, three low-level activities, for examples, 'shopping', 'cooking', and 'eating', can combine into a high-level activities, 'lunch', right? But what will happen if the sequence of the low-level activities has changes, they still form the high-level activities, 'lunch'?
b. the trigger rule for alarming
However, the method we discussed and proposed was not implemented yet; therefore, the performance (include the processing speed, accuracy like precision and recall, and efficiency) of the mechanism remains unknown.
As a thesis route student, you need to
1. design your own algorithms and mechanism of extracting the behavior pattern from a user's ADL data and identifying and clustering the patterns; therefore, you may need to
a. design structure to store ADL patterns;
b. design methods to extract ADL patterns (via using existing data mining techniques, especially time-series and sequential ways, or revising/integrating existing data mining techniques include my other student Mr. Dirksen Liu's work at http://maiga.athabascau.ca/publication/Journal-2011-JIT-Next-Stop.pdf);
c. design methods to cluster ADL patterns (via using association rules, case-based reasoning, and/or pattern comparison, etc.);
d. find cluster's generalized pattern; and,
e. compare the input pattern with either generalized cluster patterns or other people's ADL patterns.
2. evaluate the performance (include process speed, accuracy like precision and recall rate, and efficiency) of your algorithms and mechanism;
3. compare your algorithms and mechanism with other existing approaches (i.e., you also need to implement Mr. Lam and other researchers' approach); and,
4. implement a simulator which includes a floor plan and points out the user's position in the house according to the feed-in ADL data (record by record) to visualize a user's daily behavior and to show extracted patterns and its detailed information as well as to generate alarms from time to time; with this simulator, experts can be invited to assess the usability of the mechanism and to give you their thoughts about your research.
As a project route student, you need to
1. implement Mr. Lam and other researchers' approach in modules; you may need to refine or revise Mr. Lam's algorithms and mechanism as he may not consider it deeper enough;
2. evaluate the performance (include process speed, accuracy like precision and recall rate, and efficiency) of Mr. Lam algorithms and mechanism (or the revised one you made); and,
3. compare the algorithms and mechanism (Mr. Lam's or your revision) with other existing approaches. |
