Accurate prediction of urban traffic flows brings enormous advantages to big cities. Therefore, many urban traffic flow predictors have been developed in recent years. Urban traffic flow predictors aim to identify complex mobility patterns and capture urban traffic flow characteristics from large-scale historical datasets. Afterward, trained models are used to predict the future traffic volume in terms of the number of moving objects (e.g., vehicles). Convolutional Neural Networks (CNN) and other deep learning approaches are brilliant choices because of their ability to learn Spatio-temporal dependencies. Nevertheless, the extensive set of hyper-parameters tends to make these neural networks overly complex and challenging to design. In this work, we introduce RL-CNN, a framework based on Reinforcement Learning whose objective is to autonomously discover highperformance CNN architectures for the given traffic prediction task without human intervention. We examine the proposed RL-CNN model as a traffic flow estimator on a real-world and large-scale vehicular network dataset. We observe improvements of 5% - 10% in the average traffic flow prediction accuracy over the state-of-art approaches.

Predicting traffic flow in large cities is beneficial for a wide range of applications, including vehicle navigation services, vehicle routing, and traffic congestion management. In this scenario, deep learning approaches such as Recurrent Neural Networks (RNN) and its variant Long Short Term Memory (LSTM) are excellent alternatives due to their ability to learn long-term dependencies. However, these neural networks only learn the temporal traffic information for each trajectory (moving object), failing to take advantage of spatial information shared by neighboring trajectories. This paper introduces MTL-LSTM (Multi-Task Learning-based LSTM) traffic flow estimator, which attempts to explore both temporal and spatial dependencies among adjacent trajectories. Specifically, we employ LSTM predictors with the MTL approach to explore traffic flow patterns across urban trajectories. To examine the proposed model, we predict traffic flow in Porto's city using a data set from buses and taxies. Our experiments show improvements of 10% to 15% over the state-of-the-art.

Trajectory and traffic flow prediction will play an essential role in Intelligent Transportation Systems (ITS) to enable a whole new set of applications ranging from traffic management to infotainment applications. In this scenario, deep learning approaches such as Recurrent Neural Networks (RNN) and its variant Long Short Term Memory (LSTM) are excellent alternatives due to their ability to learn spatiotemporal dependencies. However, these neural networks tend to be over-complex and hard to design due to the broad set of hyper-parameters. We propose an automated framework to predict future trajectories and traffic flows in urban areas without human interventions. We employ Reinforcement Learning (RL) and Transfer Learning (TL) to generate high-performance LSTM predictors, which is referred as RL-LSTM. In addition, we introduce HERITOR (High ordE r tR affI c convoluTiOn R 1-lstm), a novel deep learning algorithm for traffic flow prediction. Specifically, HERITOR attempts to capture pure spatiotemporal features of urban traffic. The extracted features are fed into the RL-LSTM to realize a high performance LSTM for traffic flow prediction. We examine the proposed trajectory and traffic flow predictors on two real-world, large-scale datasets and observe consistent improvements of 15% - 25% over the state-of-the-art.

Future mobile networks will enable the massive deployment of mobile multimedia applications anytime and anywhere. In this context, mobility management schemes, such as handover and proactive multimedia service migration, will be essential to improve network performance. In this article, we propose a proactive mobility management approach based on group user trajectory prediction. Specifically, we introduce a mobile user trajectory prediction algorithm by combining the Long-Short Term Memory networks (LSTM) with Reinforcement Learning (RL) to automate the model training procedure. We further develop a group user trajectory predictor to reduce prediction calculation overheads of users with similar movement patterns. To validate the impact of the proposed mobility management approach, we present a virtual reality (VR) service migration scheme built on the top of the proactive handover mechanism that benefits from trajectory predictions. Experiment results validate our predictor's outstanding accuracy and its impacts on enhancing handover and service migration performance to provide quality of service assurance.

The growing ubiquity of smart-phones equipped with built-in sensors and global positioning system (GPS) has resulted in the collection of large volumes of mobility data without the need of any additional devices. The large size of heterogeneous mobility data gives rise to rapid development of location-based services (LBSs). The predictability of mobile users' behavior is essential to enhance LBSs. To predict human mobility, many techniques have been proposed. However, existing techniques require good data quality to guarantee optimal performance. In this paper, we proposed a hybrid Markov chain to predict mobile users' future locations. Our model constantly adapts to available user trace quality to select either the first order or the second order Markov chain. Compared to existing solutions, our model is adaptive to discrete gaps in data trace. To help us understanding complex user behaviors, we have also proposed a technique benefiting both temporal and spatial parameters to extract Zone of Interests (ZOIs). To evaluate the algorithm's performance, we use a real-life dataset from the Nokia Mobile Data Challenge (MDC) collected around Lake Geneva region from 180 users. We found a satisfactory future user location prediction accuracy of 70 - 84%.

The prevalence of smartphones equipped with global positioning system has enabled researchers to excavate users mobility patterns in the cities. The knowledge of users' behavior, such as their locations, plays a significant role in location-based services, resource management, logistic administration and urban planning. To understand complex behavior of humans we utilize spatio-temporal analysis on collected geo-location points to exploit Individual Zone of Interests in urban areas. In addition, we designed a hybrid Markov chain model to forecast future locations of pedestrians. Compared to existing mobility prediction methodologies, our predictor can adapt it's behavior constantly based on the quality of existing traced data to switch between first-order or second-order Markov chain. Moreover, we propose a model to predict city area congestion. The model predicts the number of users in a specific area of a city by discovering the regular mobility patterns of a group of users. We conducted comprehensive empirical experiments using a real-life dataset, namely the Mobile Data Challenge dataset, which was collected in the city of Lausanne in Switzerland with around 180 participants. We found a satisfactory user future location prediction accuracy of 70-84% and area congestion prediction accuracy of 65-73% for the users.

Content prefetching brings contents close to end users before their explicit requests to reduce the content retrieval time, which is crucial for mobile scenarios, such as vehicular ad-hoc networks (VANETs). In order to make intelligent prefetching decisions, three questions have to be answered: which content should be prefetched, when and where it should be prefetched. This paper answers these questions by proposing a vehicle mobility prediction-based over-the-top (OTT) content prefetching solution. We proposed a vehicle mobility prediction module to estimate the future connected roadside units (RSUs) using data traces collected from a real-world VANET testbed deployed in the city of Porto, Portugal. We designed a multi-tier caching mechanism with an OTT content popularity estimation scheme to forecast the content request distribution. We implemented a learning-based algorithm to proactively prefetch the user content to VANET edge caching at RSUs. We implemented a prototype using Raspberry Pi emulating RSU nodes to prove the system functionality. We also performed large-scale OpenStack experiments to validate the system scalability. Extensive experiment results prove that the system can bring benefits for both end-users and OTT service providers, which help them to optimize network resource utilization and reduce bandwidth consumption.