Privacy preserving vehicular trajectory prediction

Location based services (LBS) provide many valuable and important services for end users but reveal information about personal location to potentially untrustworthy service providers which could pose privacy concerns. The information about exact path followed by a person is highly personal and can be used to uniquely identify a person. Such information is known as personally identifiable information (PII). Disclosing such information can be considered as breach of privacy. Data needs to be transformed in such a way that either PII is removed or cannot be inferred preserving the accuracy of inferences at the same time. Differential Privacy [1] is a widely used techniques for removing PII from the data. It aims to provide means to maximize the accuracy of queries from statistical databases while minimizing the chances of identifying its records. This work aims at studying the impact of transforming the data on its utility. We have used two differentially privacy techniques Laplace perturbation and sliding moving average noise reduction on trajectories of taxi [2]. This dataset is used to predict the final destination (latitude and longitude) of taxi trips. These differential privacy techniques maintain the utility of this dataset along with the preservation of the trajectory privacy of individuals.