Application of PageRank for Anomaly Detection in Movements
https://www.hindawi.com/journals/ijta/2019/8612021/
We have talked about Google’s PageRank algorithm in class as a way to rank web pages based on the number and quality of links directed into each page. In this research article, the authors present an application of PageRank to a different field – monitoring movement patterns of older adults in their homes. The purpose of doing so is to watch for anomalies in their health by comparing daily sensor data to a normal pattern computed using PageRank.
First, a base graph is made representing a normal pattern of movement. In a graphical model, rooms around the house are depicted as nodes, and there are directed edges representing the directions of motion between each of the nodes. PageRank is then applied to this graph to rank the rooms/hallways in order of how often the person visits them and what preferences they tend to have. They collect and assign a probability distribution that the older adult can visit a selected room/node from any other room/node. The probability of them being at a given room is calculated as a PageRank value in a similar way to what we did in class – the value is based on the sum of the probabilities of the older adult being in the rooms that point to the current room at the previous instant, and also based on the number of edges pointing out of each of those adjacent rooms. So basically, the value for each room is dependent on how many rooms lead to it and the probabilities associated with those rooms. The authors took the PageRank method a step further, and instead of splitting the PageRank values of nodes equally over its outgoing links, they added a transition weight factor to normalize their results.
The authors’ hypothesis was that they would be able to detect anomalous activity in movement pattern by looking for changes from the base pattern in the nominal PageRank. For example, you can track the older adult’s movement with sensors during an observation period and calculate their room preferences during this period. Then, the PageRank algorithm could be re-applied to that data to check if there is a change in the ranking of the rooms. This would indicate that their tendencies during this period were atypical.
Being able to detect deviations in daily movement patterns can provide insight to physicians about the risk of frailty, and can help detect health problems, both physical (ex: incontinence) and mental (ex: Alzheimer’s disease). Finding a successful way to monitor this can allow older adults to maintain an active lifestyle while still remaining in their private home.
I thought that this research article was interesting because it took the PageRank method, which was originally intended for ranking web pages for a search engine, and showed how it could be used in a very different scenario. Seeing the benefits it could provide to health monitoring opens up the possibilities of other applications where PageRank can be useful.