Human Mobility Support for Personalized Data Offloading

Predictive models for human mobility have important applications in many fields including traffic control, ubiquitous computing, and contextual advertisement. The predictive performance of models in literature varies quite broadly, from over 90% to under 40%. In this work we study which underlying factors - in terms of modeling approaches and spatio-temporal characteristics of the data sources - have resulted in this remarkably broad span of performance reported in the literature. Specifically we investigate which factors influence the accuracy of next-place prediction, using a high-precision location dataset of more than 400 users observed for periods between 3 months and one year. We show that it is much easier to achieve high accuracy when predicting the time-bin location than when predicting the next place. Moreover, we demonstrate how the temporal and spatial resolution of the data have strong influence on the accuracy of prediction. Finally we reveal that the exploration of new locations is an important factor in human mobility, and we measure that on average 20-25% of transitions are to new places, and approx. 70% of locations are visited only once. We discuss how these mechanisms are important factors limiting our ability to predict human mobility.

The prediction of human and vehicle mobility in a city is becoming an attracting field. This topic attracts researchers from broad fields including behavioral sciences, where understanding the complexity of the human mobility behavior is one of the hot topic, and also to the industrial partners, who apply such results to many beneficial applications. Recent progress to sensing human mobility via smartphones is boosting this trend. However, due to the complexity and context-dependence of human behavior and the incompleteness and noise of geospatial data collecting from various sensors, the prediction of human and vehicle mobility is still far from solved. This workshop aimed at collecting contributions on the cutting-edge studies in human mobility description, modeling, intelligent computational method which can advance the human and vehicle prediction research. Potential topics included, but were not limited to 1) The next location prediction of individual mobility, 2) The crowd or population mobility prediction, 3) Dynamics of pedestrians, 4) commute flow and migration flow, 5) Traffic congestion, road usage forecast and optimal vehicle routing, 6) Social event forecast using geospatial data, 7) Novel agent mobility simulators, and 8) Case studies of mobility estimation in academia as well as in industrial field.

Mobility prediction plays important roles in many fields. For example, tourist companies would like to know the characteristics of their customer movements so that they could design appropriate advertising strategies; sociologists has made many research on migration to try to find general features in human mobility; polices also analyze human movement behaviors to seek criminals. Thus, for location-based social networks, mobility prediction is an important task. This study proposes a mobility prediction model, which can be used to predict the user (human) mobility. The proposed approach is conducted from three characteristics: (1) regular movement in human mobility, (2) the influence of relationships on social networks, (3) other features (in this work, we consider “hot regions” where attract more people coming to there). To validate the proposed approach, three datasets including over 500,000 check-ins which are collected from two location-based social networks, namely Brightkite and Gowalla, are used for the experiments. Results show that the proposed model significantly improves the prediction accuracy, thus, this approach could be promising for mobility prediction, especially for location-based social networks.

In recent decades, the frequency and intensity of natural disasters has increased significantly, and this trend is expected to continue. Therefore, understanding and predicting human behavior and mobility during a disaster will play a vital role in planning effective humanitarian relief, disaster management, and long-term societal reconstruction. However, such research is very difficult to perform owing to the uniqueness of various disasters and the unavailability of reliable and large-scale human mobility data. In this study, we collect big and heterogeneous data (e.g., GPS records of 1.6 million users 1 over 3 years, data on earthquakes that have occurred in Japan over 4 years, news report data, and transportation network data) to study human mobility following natural disasters. An empirical analysis is conducted to explore the basic laws governing human mobility following disasters, and an effective human mobility model is developed to predict and simulate population movements. The experimental results demonstrate the efficiency of our model, and they suggest that human mobility following disasters can be significantly more predictable and be more easily simulated than previously thought.

This chapter summarizes the background and motivation for why people need to make every effort to predict network traffic and user mobility. To address all the technical aspects related to traffic and mobility predictions with the emphasis on the machine learning based solutions, the chapter first presents the problem formation and a brief overview of existing prediction methods in terms of modelling, characterization, complexity, and performance. It presents some deep learning based schemes for traffic and mobility prediction. The chapter also introduces random connectivity long short‐term memory – a model that reduces computational cost by randomly removing some neural connections – and its performance in traffic prediction. Then it shows three long short‐term memory‐based user‐mobility prediction schemes, two of which take into account the spatial dependence on the user's movement trajectory, thus combining convolutional neural networks to expect to achieve better prediction performance.

With the development of the Internet, the issue of cyberbullying on social media has gained significant attention. Cyberbullying is often expressed in text. Methods of identifying such text via machine learning have been growing, most of which rely on the extraction of part-of-speech (POS) tags to improve their performance. However, the current study only arbitrarily used part-of-speech labels that it considered reasonable, without investigating whether the chosen part-of-speech labels can better enhance the effectiveness of the cyberbullying detection task. In other words, the effectiveness of different part-of-speech labels in the automatic cyberbullying detection task was not proven. This study aimed to investigate the part of speech in statements related to cyberbullying and explore how three classification models (random forest, naïve Bayes, and support vector machine) are sensitive to parts of speech in detecting cyberbullying. We also examined which part-of-speech combinations are most appropriate for the models mentioned above. The results of our experiments showed that the predictive performance of different models differs when using different part-of-speech tags as inputs. Random forest showed the best predictive performance, and naive Bayes and support vector machine followed, respectively. Meanwhile, across the different models, the sensitivity to different part-of-speech tags was consistent, with greater sensitivity shown towards nouns, verbs, and measure words, and lower sensitivity shown towards adjectives and pronouns. We also found that the combination of different parts of speech as inputs had an influence on the predictive performance of the models. This study will help researchers to determine which combination of part-of-speech categories is appropriate to improve the accuracy of cyberbullying detection.

BackgroundAccurate prediction of delirium in the intensive care unit (ICU) may facilitate efficient use of early preventive strategies and stratification of ICU patients by delirium risk in clinical research, but the optimal delirium prediction model to use is unclear. We compared the predictive performance and user convenience of the prediction model for delirium (PRE-DELIRIC) and early prediction model for delirium (E-PRE-DELIRIC) in ICU patients and determined the value of a two-stage calculation.MethodsThis 7-country, 11-hospital, prospective cohort study evaluated consecutive adults admitted to the ICU who could be reliably assessed for delirium using the Confusion Assessment Method-ICU or the Intensive Care Delirium Screening Checklist. The predictive performance of the models was measured using the area under the receiver operating characteristic curve. Calibration was assessed graphically. A physician questionnaire evaluated user convenience. For the two-stage calculation we used E-PRE-DELIRIC immediately after ICU admission and updated the prediction using PRE-DELIRIC after 24 h.ResultsIn total 2178 patients were included. The area under the receiver operating characteristic curve was significantly greater for PRE-DELIRIC (0.74 (95% confidence interval 0.71–0.76)) compared to E-PRE-DELIRIC (0.68 (95% confidence interval 0.66–0.71)) (z score of − 2.73 (p < 0.01)). Both models were well-calibrated. The sensitivity improved when using the two-stage calculation in low-risk patients. Compared to PRE-DELIRIC, ICU physicians (n = 68) rated the E-PRE-DELIRIC model more feasible.ConclusionsWhile both ICU delirium prediction models have moderate-to-good performance, the PRE-DELIRIC model predicts delirium better. However, ICU physicians rated the user convenience of E-PRE-DELIRIC superior to PRE-DELIRIC. In low-risk patients the delirium prediction further improves after an update with the PRE-DELIRIC model after 24 h.Trial registrationClinicalTrials.gov, NCT02518646. Registered on 21 July 2015.

Abstract. Benchmarking model performance across large samples of catchments is useful to guide model selection and future model development. Given uncertainties in the observational data we use to drive and evaluate hydrological models, and uncertainties in the structure and parameterisation of models we use to produce hydrological simulations and predictions, it is essential that model evaluation is undertaken within an uncertainty analysis framework. Here, we benchmark the capability of several lumped hydrological models across Great Britain by focusing on daily flow and peak flow simulation. Four hydrological model structures from the Framework for Understanding Structural Errors (FUSE) were applied to over 1000 catchments in England, Wales and Scotland. Model performance was then evaluated using standard performance metrics for daily flows and novel performance metrics for peak flows considering parameter uncertainty. Our results show that lumped hydrological models were able to produce adequate simulations across most of Great Britain, with each model producing simulations exceeding a 0.5 Nash–Sutcliffe efficiency for at least 80 % of catchments. All four models showed a similar spatial pattern of performance, producing better simulations in the wetter catchments to the west and poor model performance in central Scotland and south-eastern England. Poor model performance was often linked to the catchment water balance, with models unable to capture the catchment hydrology where the water balance did not close. Overall, performance was similar between model structures, but different models performed better for different catchment characteristics and metrics, as well as for assessing daily or peak flows, leading to the ensemble of model structures outperforming any single structure, thus demonstrating the value of using multi-model structures across a large sample of different catchment behaviours. This research evaluates what conceptual lumped models can achieve as a performance benchmark and provides interesting insights into where and why these simple models may fail. The large number of river catchments included in this study makes it an appropriate benchmark for any future developments of a national model of Great Britain.

Quantitative analysis of polycyclic aromatic hydrocarbons in soil by infrared spectroscopy combined with hybrid variable selection strategy and partial least squares

Multi-city human mobility prediction is critical for GIScience and urban computing. However, pronounced spatiotemporal heterogeneity and training instability on large-scale, multi-city datasets restrict model generalization. To address these challenges, we propose the Human Mobility Predictor (HuMob Predictor), a novel framework that enhances multi-city mobility prediction. Our model introduces a multi-level encoding module that captures cross-city heterogeneity and within-city spatial relationships. City encodings capture macro-level characteristics unique to each city, while absolute spatial encodings preserve geographic proximity across grid cells. By combining these encodings, HuMob Predictor learns universal mobility representations while retaining city-specific features. To stabilize training across cities, we adopt an incremental training strategy that gradually increases prediction difficulty, significantly improving convergence and cross-city generalization. Experiments on the GISCUP 2025 multi-city datasets demonstrate that HuMob Predictor achieves superior performance in individual mobility prediction.

Prognostication for patients with cancer is important for clinical planning and management, but remains challenging given the large number of factors that can influence outcomes. As such, there is a need to identify features that can robustly predict patient outcomes. We evaluated 8608 patient tumor samples across 16 cancer types from The Cancer Genome Atlas and generated distinct survival classifiers for each using clinical and histopathological data accessible to standard oncology workflows. For cancers that had poor model performance, we deployed a random-forest-embedded sequential forward selection approach that began with an initial subset of the 15 most predictive clinicopathological features before sequentially appending the next most informative gene as an additional feature. With classifiers derived from clinical and histopathological features alone, we observed cancer-type-dependent model performance and an area under the receiver operating curve (AUROC) range of 0.65 to 0.91 across all 16 cancer types for 1- and 3-year survival prediction, with some classifiers consistently outperforming those for others. As such, for cancers that had poor model performance, we posited that the addition of more complex biomolecular features could enhance our ability to prognose patients where clinicopathological features were insufficient. With the inclusion of gene expression data, model performance for 3 select cancers (glioblastoma, stomach/gastric adenocarcinoma, ovarian serous carcinoma) markedly increased from initial AUROC scores of 0.66, 0.69, and 0.67 to 0.76, 0.77, and 0.77, respectively. As a whole, this study provides a thorough examination of the relative contributions of clinical, pathological, and gene expression data in predicting overall survival and reveals cancer types for which clinical features are already strong predictors and those where additional biomolecular information is needed.

Background Reasons for the relatively poor performance of bleeding prediction models are not well understood but may relate to differences in predictors for various anatomical sites of bleeding.Methods We pooled individual participant data from four randomized controlled trials of antithrombotic therapy in patients with coronary and peripheral artery diseases, embolic stroke of undetermined source (ESUS), or atrial fibrillation. We examined discrimination and calibration of models for any major bleeding, major gastrointestinal (GI) bleeding, and intracranial hemorrhage (ICH), according to the time since initiation of antithrombotic therapy, and indication for antithrombotic therapy.Results Of 57,813 patients included, 1,948 (3.37%) experienced major bleeding, including 717 (1.24%) major GI bleeding and 274 (0.47%) ICH. The model derived to predict major bleeding at 1 year from any site (c-index, 0.69, 95% confidence interval [CI], 0.68–0.71) performed similarly when applied to predict major GI bleeding (0.71, 0.69–0.74), but less well to predict ICH (0.64, 0.61–0.69). Models derived to predict GI bleeding (0.75, 0.74–0.78) and ICH (0.72, 0.70–0.79) performed better than the general major bleeding model. Discrimination declined over time since the initiation of antithrombotic treatment, stabilizing at approximately 2 years for any major bleeding and major GI bleeding and 1 year for ICH. Discrimination was best for the model predicting ICH in the ESUS population (0.82, 0.78–0.92) and worst for the model predicting any major bleeding in the coronary and peripheral artery disease population (0.66, 0.65–0.69).Conclusion Performance of risk prediction models for major bleeding is affected by site of bleeding, time since initiation of antithrombotic therapy, and indication for antithrombotic therapy.

Classical regression is a supervised task that uses target output to guide the modeling. Generally, the original input contains both output relevant and irrelevant information, whereas the latter may decrease the predictive performance to a certain extent. Without prior information about what information contributes to the prediction, the predictive performance of regression models can be improved based on such output relevant information as input. Thus, a joint autoencoder (JAE) combining the supervised and unsupervised mechanisms is proposed to learn output relevant features from the original input. In this way, both predictive and reconstructive performance are considered to learn the essential characteristics and avoid poor generalized performance on untrained instances. Meanwhile, hierarchy representations can be learned by successive JAEs with a local parameter embedding strategy, which is presented to preserve the predictive performance in this structure. In the experiments, the predictive performance and robustness of the proposal are verified on nine datasets with different sample sizes.

The objective of this study is to evaluate the predictive performance of several models to predict drug clearance in preterm and term neonates. Five models using different types of allometric and linear models were developed. Two sets of data were used to develop these models (data from preterm neonates to adults and data from preterm and term neonates). Models were also developed with (normalized to 70kg) or without body weight normalization (body weight 1kg). From the literature, clearance values for four drugs from neonates to adults were obtained. External data were used to evaluate the predictive performance of these models in preterm and term neonates. The results of the study indicated that (1) normalization to a standard body weight had no impact on the predictive performance of the models, (2) the model developed from preterm neonates to adults using fixed exponent 0.75 provided inaccurate estimate (overestimation) of drug clearance in neonates, (3) a far superior prediction of clearance was observed with the model when the exponents of allometry were estimated than the model using exponent 0.75, (4) linear models with the exception of the model with intercept provided comparable results to the estimated exponent model and were superior in their predictive performance to the model using exponent 0.75, and (5) when the models were developed from neonate data, the predictive performance of all models were similar. Overall, the study indicated that body weight normalization had no impact on the performance of model prediction, the exponents of allometry in pharmacostatistical models should be estimated rather than fixed, and more studies are needed to evaluate the suitability of linear models for the prediction of drug clearance in neonates.

Human mobility is the foundation of urban dynamics and its prediction significantly benefits various downstream location-based services. Nowadays, while deep learning approaches are dominating the mobility prediction field where various model architectures/designs are continuously updating to push up the prediction accuracy, there naturally arises a question: whether these models are sufficiently good to reach the best possible prediction accuracy? To answer this question, predictability study is a method that quantifies the inherent regularities of the human mobility data and links the result to that limit. Mainstream predictability studies achieve this by analyzing the individual trajectories and merging all individual results to obtain an upper bound. However, the multiple individuals composing the city are not totally independent and the individual behavior is heavily influenced by its implicit or explicit surroundings. Therefore, the collective factor should be considered in the mobility predictability measurement, which has not been addressed before. This vision paper points out this concern and envisions a few potential research problems along such an individual-to-collective transition from both data and methodology aspects. We hope the discussion in this paper sheds some light on the human mobility predictability community.