Business Applications Based on Human Mobility Analysis and Travel Prediction
Carlos Andre Reis Pinheiro, Data Scientist EMC2, FGV, KU Leuven– January 2015
Human mobility analysis reveals relevant knowledge about subscribers’ behavior as well as urban planning, traffic forecasting or a better understanding about spread diseases. Human mobility studies based on mobile phone data may disclose an approximation of the human motion within particular geographic areas such as great metropolitan areas, big cities, states and countries, including migration trends between them.
Mobile phone data provides a set of information about the caller and called numbers, assigning to them the cell towers spread out through the urban area. These cell towers hand out the calls and texts made and receive by the subscribers – including the internet access, and give us some useful information about their geographic locations in specific points of time, such as the cell tower’s latitude and longitude, it’s radius of coverage, and also address information including neighborhood, district and county.
The traditional predictive models are mostly based on past events assigned to a target variable, allowing us to train a particular model in eventually predict that target. The training is based on the historical events and the correlation of them to the target class.
The models from transportation systems – gravity and radiation – are mostly based on the characteristics of the locations, in analogy to the gravitational attraction between bodies and the radiation emission between particles. It is intuitive to think about how a body with high mass can attract another body with low mass –earth and apple – and how the heat flows from one warm body to a cold one.
In the mobility and transportation phenomena, the most relevant characteristics include the population of the locations and the distance between them. Additional information may be incorporated to improve the prediction rate such as the number of jobs in one particular location, positions at universities and schools, technical courses, amount of restaurants and other leisure options, among many others. The main explanation for the trips between locations – according to the gravity and the radiation models – is that one location attracts people from another for some reason. The basic motif is that locations with higher population would attract people from locations with lower population, decreased by the cost of the travel – distance or the time to travel.
At the lowest level of prediction for instance, based on the subscriber’s frequent trajectory, it would be possible – notwithstanding the policies and regulations in relation to privacy – to foresee where and when the users would be. Suppose that a particular mobile company has the overall frequent trajectories for all subscribers over time. This company would be able to foresee that a specific user would stay at location l at the time t.
Suppose also – if possible due to the privacy restrictions – that this mobile company has partnerships with other companies or sell advertisements in its network – like banners in websites. Any company would be able then to offer something to that subscriber considering that he or she would most likely to be at location l at time t. Imagine McDonald’s or Starbucks sending a message – through that mobile company – about a offering, in a particular store (near to location l) during a specific period of time (around time t). The chances to succeed in this broadcasting would be significantly higher than it was randomly advertised.
Gravity and radiation models both provide a method to raise new insights and useful knowledge about population mobility behavior. These models stand on a solid foundation and can accurately reproduce mobility patterns varying from migration paths to commutes trajectories.
The gravity model usually requires some parameters to define constraints on the generation and attractions of flow such as distance and cost of travel. It states that the commuting between two locations is proportional to the product of the population of these two places and inversely proportional to a power law of the distance between these two places. On the other hand, the radiation model requires just the spatial distribution population as input, and no further adjustable parameters. It basically depends only on the population in the origin, the population in the destination and on the population in a circle where the center is the origin with a radius as the distance between the origin and the destination (excluding the populations in the origin and destination locations).
The gravity model assumes that individuals are attracted to other locations as a function of the distance between these two places associated to the cost of the travel distance between them. The radiation model assumes that individuals are attracted by the nearest locations rather than the farther locations with more opportunities due to the hypothesis of limited resources of mobility and high cost of travels. In both models the travel distance is a crucial factor in decision making of users in commuting between two locations.
In addition to the predictive models, when analyzing the mobility behavior over time we may also identify some particularities of the human’s motion progress and raise some insights from it.
Mobile carriers, for instance, might analyze how evolves their subscribers’ mobility over time and based on that knowledge optimize the communications’ network, including trends in terms of amount of visits the cell towers would have over time, defining then alternative paths and routes and contingency approaches in expected events – such as shows, sports and special dates – as well as in unexpected ones – such as flood, torrents, public services outage, among others.
In the public sector, the health care department can use the vector of movements over time to better understand how some particular diseases spread out through the city. Analogously, transportation agencies may compare the vectors of movement over time with the current possible routes in the city and increase their knowledge about eventual solutions for the public transportation planning.
There are many benefits that can be ranked by using the outcomes from the human mobility analysis, for both public sector and private industry. The understanding of how people move throughout great cities can reveal relevant information for better plan many different subjects.