Multiple Systems Combination to Improve Human Mobility Prediction

Effective human mobility modeling and prediction constitute the core prerequisites for various location-based applications. To encourage research in this direction, the ACM SIGSPATIAL Cup 2025 posed the challenge of predicting human mobility trajectories from a sparse multi-city dataset. This paper presents our solution, MoBERT, a BERT-like model that adapts and leverages mobility semantics with additional direction and distance prediction, providing supplementary supervision signals for robust feature learning. MoBERT models are trained in stages through curriculum learning, where augmented trajectories are ordered by increasing mobility entropy for training with progressively increasing difficulty. The final score of our method is 0.14609, as measured by average GEO-BLEU distances across four cities. Finally, we analyze the results and discuss insights from our approach.

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.

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.

We propose CLIFT ( C ross-City Lif estyle Pattern T ransfer for Human Mobility Prediction), a novel framework that enhances human mobility prediction by integrating general lifestyle patterns shared across cities with city-specific mobility patterns. Accurate human mobility prediction in urban environments is critical for transportation planning, marketing strategies, and disaster response. However, most existing deep learning approaches use only single-city data and exhibit significant performance degradation in small cities with limited training data. These limitations motivate methods that jointly leverage cross-city behavioral patterns and city-specific mobility characteristics. CLIFT addresses this challenge through dual complementary encoders: one captures general lifestyle patterns shared across cities, and the other captures city-specific mobility patterns; their outputs are integrated with a Transformer-based mobility predictor (LP-BERT). This architecture enables the model to jointly capture cross-city transferable behavioral patterns and city-specific mobility characteristics. We evaluated the effectiveness of CLIFT through experiments on the multi-city human mobility dataset LYMob-4Cities, comparing its performance with both single-city and multi-city deep learning-based methods. On average, CLIFT improved GEOBLEU and Top-1 accuracy by 11.1% and 10.6% over the single-city baseline, and by 5.0% and 7.9% over the multi-city baseline, respectively. Furthermore, CLIFT outperformed the top-ranked teams in the international competition, Human Mobility Prediction Challenge 2024 , demonstrating superior predictive performance under the same dataset and task setting.

Human mobility prediction is of great importance in a wide range of modern applications in different fields such as personalized recommendation systems, the fifth-generation (5G) mobile communication systems, and so on. Generally, the prediction goal varies from different application scenarios. For the applications of 5G network including resource allocation and mobility management, it is essential to predict the positions of mobile users in the near future from dozens of seconds to a few minutes so as to make preparation in advance, which is actually a trajectory prediction problem. In this paper, with the particular focus on multi-user multi-step trajectory prediction, we first design a basic deep learning-based prediction framework, where the long short-term memory (LSTM) network is directly applied as the most critical component to learn user-specific mobility pattern from the user's historical trajectories and predict his/her movement trends in the future. Motivated by the related findings after testifying and analyzing this basic framework on a model-based dataset, we extend it to a region-oriented prediction scheme and propose a multi-user multi-step trajectory prediction framework by further incorporating the sequence-to-sequence (Seq2Seq) learning. The experimental results on a realistic dataset demonstrate that the proposed framework has significant improvements on generalization ability and reduces error-accumulation effect for multi-step prediction.

As an initial research in human mobility, human mobility prediction can be done by using a time-series predictor algorithm, one of which is ARIMA. ARIMA is short of the integrated moving-average autoregressive. The order of the ARIMA model is represented by the ARIMA symbol (p, d, q), where p is the order of the autoregressive part, d is the order of differencing and q is the order of moving-average process. The research regarding the application of human mobility conducted through five phases, including data collection, data pre-processing, data model building, data prediction and data evaluation. We conducted three times of experiments with different parameters. We defined different value for D, Seasonality, MALags, SMALags and Variance. Based on experiment as conclusion of this research obtained that the best parameter values to get the best MAPE Longitude and MAPE Latitude are Constant = 0, D= 1, Seasonality= 12, MALags = 1, SMALags=12 with MAPE Lon: 0.037433% and MAPE Lat: 0.11632%

The quality of life for people in urban regions can be improved by predicting urban human mobility and adjusting urban planning accordingly. In this study, we compared several possible variables to verify whether a gravity model (a human mobility prediction model borrowed from Newtonian mechanics) worked as well in inner-city regions as it did in intra-city regions. We reviewed the resident population, the number of employees, and the number of SNS posts as variables for generating mass values for an urban traffic gravity model. We also compared the straight-line distance, travel distance, and the impact of time as possible distance values. We defined the functions of urban regions on the basis of public records and SNS data to reflect the diverse social factors in urban regions. In this process, we conducted a dimension reduction method for the public record data and used a machine learning-based clustering algorithm for the SNS data. In doing so, we found that functional distance could be defined as the Euclidean distance between social function vectors in urban regions. Finally, we examined whether the functional distance was a variable that had a significant impact on urban human mobility.

Human mobility prediction is a challenging and crucial task for fostering wide spectrum of location based applications, namely, traffic planning, travel-plan recommendation etc. However, modeling human movement behaviour and predicting next location are non-trivial tasks. We consider the human mobility pattern modeling and predicting next location problem for semantic trajectory data, wherein sparse GPS records are associated with other contextual information. Intuitively, human moves with an intent and the confluence of movement history (GPS traces) and contextual information may help in capturing people's intents at a fine granularity and has potential to improve the efficacy of location prediction problem. In this paper, a hierarchical and layered hidden markov model (HMM) framework for mobility prediction from sparse trajectories has been proposed. Experiment on real-life mobility dataset of Nokia Mobile Data Challenge (MDC) demonstrates the efficacy of the proposed framework even when with sparse GPS traces.

Most Graph Neural Networks (GNNs) follow the message-passing scheme. Residual connection is an effective strategy to tackle GNNs' over-smoothing issue and performance reduction issue on non-homophilic networks. Unfortunately, the coarse-grained residual connection still suffers from class-imbalanced over-smoothing issue, due to the fixed and linear combination of topology and attribute in node representation learning. To make the combination flexible to capture complicated relationship, this paper reveals that the residual connection needs to be node-dependent, layer-dependent, and related to both topology and attribute. To alleviate the difficulty in specifying complicated relationship, this paper presents a novel perspective on GNNs, i.e., the representations of one node in different layers can be seen as a sequence of states. From this perspective, existing residual connections are not flexible enough for sequence modeling. Therefore, a novel node-dependent residual connection, i.e., Long Short-Term Graph Memory Network (LSTGM) is proposed to employ Long Short-Term Memory (LSTM), to model the sequence of node representation. To make the graph topology fully employed, LSTGM innovatively enhances the updated memory and three gates with graph topology. A speedup version is also proposed for effective training. Experimental evaluations on real-world datasets demonstrate their effectiveness in preventing over-smoothing issue and handling networks with heterophily.

Understanding and predicting mobility are essential for the design and evaluation of future mobile edge caching and networking. Consequently, research on human mobility prediction has drawn significant attention in the last decade. Employing information-theoretic concepts and machine learning methods, earlier research has shown evidence that human behavior can be highly predictable. Whether high predictability manifests itself for different modes of device usage, across spatial and temporal dimensions is still debatable. Despite existing studies, more investigations are needed to capture intrinsic mobility characteristics constraining predictability, to explore more dimensions (e.g. device types) and spatiotemporal granularities, especially with the change in human behavior and technology. We investigate practical predictability of next location visitation across three different dimensions: device type, spatial granularity and temporal spans using an extensive longitudinal dataset, with fine spatial granularity (AP level) covering 16 months. The study reveals device type as an important factor affecting predictability. Ultra-portable devices such as smartphones have "on-the-go" mode of usage (and hence dubbed "Flutes"), whereas laptops are "sit-to-use" (dubbed "Cellos"). The goal of this study is to investigate practical prediction mechanisms to quantify predictability as an aspect of human mobility modeling, across time, space and device types. We apply our systematic analysis to wireless traces from a large university campus. We compare several algorithms using varying degrees of temporal and spatial granularity for the two modes of devices; Flutes vs. Cellos. Through our analysis, we quantify how the mobility of Flutes is less predictable than the mobility of Cellos. In addition, this pattern is consistent across various spatio-temporal granularities, and for different methods (Markov chains, neural networks/deep learning, entropy-based estimators). This work substantiates the importance of predictability as an essential aspect of human mobility, with direct application in predictive caching, user behavior modeling and mobility simulations.

The frequency and intensity of natural disasters has significantly increased over the past decades and this trend is predicted to continue. Facing these possible and unexpected disasters, understanding and simulating of human emergency mobility following disasters will becomethe critical issue for planning effective humanitarian relief, disaster management, and long-term societal reconstruction. However, due to the uniquenessof various disasters and the unavailability of reliable and large scale human mobility data, such kind of research is very difficult to be performed. Hence, in this paper,we collect big and heterogeneous data (e.g. 1.6 million users' GPS records in three years, 17520 times of Japan earthquake data in four years, news reporting data, transportation network data and etc.) to capture and analyze human emergency mobility following different disasters. By mining these big data, we aim to understand what basic laws govern human mobility following disasters, and develop a general model of human emergency mobility for generating and simulating large amount of human emergency movements. The experimental results and validations demonstrate the efficiency of our simulation model, and suggest that human mobility following disasters may be significantly morepredictable and can be easier simulated than previously thought.

The research study is to predict human activity recognition and analyze their strength in machine learning (ML) processes to achieve precision and also to determine performance of human activities. The framework to classify the Novel Long Short Term Memory (LSTM) and Decision Tree (DT) to perform all measures. This research study used the Novel Long Short Term Memory and Decision Tree to perform the required actions to accurately identify body movements in adjacent film sequence. Recent study findings and a threshold of 0.05%, as well as 95% confidence interval and standard parameter estimates, were castoff in a study 47 samples divided into two groups calculated using the G-power of 80%. Since the novel Long Short Term Memory (LSTM) algorithm and local invariant methods have achieved an accuracy of 93.83% in predicting the activity analysis, this research needs to obtain improved accuracy to perform activity prediction by using the Decision Tree algorithm. When applied to human activity analysis, the Novel Long Short Term Memory technique was shown to have an accuracy of 90.12 percent, with a 95% confidence interval including the value 0.045 (p 0.05). Researchers found that when comparing two algorithms for analyzing human behaviour, Long Short Term Memory (LSTM) performed much better than Decision Tree (DT).

Long short-term memory - Fully connected (LSTM-FC) neural network for PM2.5 concentration prediction

Human Mobility Data in the COVID-19 Pandemic: Characteristics, Applications, and Challenges

This study applies a Long Short-Term Memory (LSTM) neural network to model daily groundwater level variations at four monitoring wells in the piedmont plain of Baoding City, Hebei Province, China. Using daily data from January 1, 2018 to August 31, 2019, the LSTM model is trained and tested under a one-step rolling prediction framework. The results demonstrate that the LSTM model accurately fits groundwater time series across various hydrogeological conditions, with testing-phase RMSE values ranging from 0.08 to 0.45 meters and R2 values exceeding 0.90. The model performs exceptionally well for both stable and fluctuating groundwater level conditions, capturing seasonal decline and recovery patterns without requiring explicit seasonal indicators. It also reveals the model’s ability to learn long-term trends and nonlinear dynamics inherent in groundwater systems. Despite its high short-term prediction accuracy, challenges remain regarding multi-step forecasting and responses to extreme events outside the training data. The study concludes that LSTM offers strong potential for groundwater simulation in data-limited environments and recommends further integration of hydro-meteorological variables for enhanced model robustness.