Secure forecasting of user activities for distributed urban applications

Abstract

Modelling human mobility is an interesting yet challenging research topic. Such mobility models can give valuable insight into user behavior. Such models can be used to forecast movement of people. Even though an interesting problem, it was not studied as widely due to lack of available mobility data. But modern communication and digital infrastructure has solved this problem. Thus, as a result, over the past decade and a half, this topic has attracted a lot of attraction. The modelling and forecasting of human mobility has widespread applications from transportation to advertisement. Such models can be used to in a collaborative manner to segment people or used in isolation to bring better services to an individual. Previous researches have presented different approaches for modelling human mobility. These range from neural networks to Markov chains. Some researchers have focused on location data while others have worked with accelerometer data. There are also recommendations to add more information to the data to understand the motive of mobility. This thesis approaches the problem of forecasting human mobility in the form of activities. GPS data is analyzed to mine information and find patterns. The forecasting is done in a twostep process. The first step is to analyze the data to identify and label activities, that are done on a routine basis. This is achieved by using an Adaptive Neuro-Fuzzy Inference System. This additional information helps understand the motive of moving from one place to another. In the second and final step the Markov Chain model is built for the movement among visited locations. The forecasting is done with respect to current time and location, keeping in view the motive of movement. The proposed system is implemented in JAVA and deployed as a combination of RESTful web services. Finally, accuracy tests are made on different datasets which show promising results.