This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

EpiLearn: A Python Library for Machine Learning in
Epidemic Modeling

Zewen Liu Department of Computer ScienceEmory University [email protected] ,  Yunxiao Li Department of Computer ScienceEmory University [email protected] ,  Mingyang Wei Department of Computer ScienceEmory University [email protected] ,  Guancheng Wan Department of Computer ScienceEmory University [email protected] ,  Max S.Y. Lau Department of Biostatistics and BioinformaticsEmory University [email protected]  and  Wei Jin Department of Computer ScienceEmory University [email protected]
(2024)
Abstract.

EpiLearn is a Python toolkit developed for modeling, simulating, and analyzing epidemic data. Although there exist several packages that also deal with epidemic modeling, they are often restricted to mechanistic models or traditional statistical tools. As machine learning continues to shape the world, the gap between these packages and the latest models has become larger. To bridge the gap and inspire innovative research in epidemic modeling, EpiLearn not only provides support for evaluating epidemic models based on machine learning, but also incorporates comprehensive tools for analyzing epidemic data, such as simulation, visualization, transformations, etc. For the convenience of both epidemiologists and data scientists, we provide a unified framework for training and evaluation of epidemic models on two tasks: Forecasting and Source Detection. To facilitate the development of new models, EpiLearn follows a modular design, making it flexible and easy to use. In addition, an interactive web application is also developed to visualize the real-world or simulated epidemic data. Our package is available at https://github.com/Emory-Melody/EpiLearn.

Epidemiology, Epidemic models, Machine learning, Neural networks, Python
††journalyear: 2024††copyright: rightsretained††conference: 7th epiDAMIK ACM SIGKDD International Workshop on Epidemiology meets Data Mining and Knowledge Discover; August 26, 2024; Barcelona, Spain††booktitle: Proceedings of the 7th epiDAMIK ACM SIGKDD International Workshop on Epidemiology meets Data Mining and Knowledge Discovery, August 26, 2024, Barcelona, Spain

1. Introduction

Data mining in epidemiology is a crucial subject in the healthcare domain, garnering increasing attention in recent years due to the COVID-19 outbreak  (pare2020modeling, 1, 2). A key focus is the development of computational methods in epidemic modeling, which incorporate disease transmission mechanisms to provide insights into changing demographic health states. The diversity of data involved in epidemic modeling necessitates a broad range of tasks, including epidemic forecasting  (zheng2021hierst, 3), simulation  (chao2010flute, 4), source detection  (sha2021source, 5), intervention strategies  (song2020reinforced, 6), and vaccination  (jhuneffective, 7).

Traditionally, knowledge-driven approaches like mechanistic models (e.g., SIR, SIS, and SEIR (he2020seir, 8)) have been employed for epidemic modeling. These methods utilize differential equations to explicitly model relationships among population groups in different states, e.g. Suspected, Infected, and Recovered, and have demonstrated promising performance. However, over the past decade, the rapid progress in machine learning and deep learning (ResNet_CVPR16, 9, 10, 11, 12, 13, 14, 15) has paved the way for incorporating these techniques into epidemic modeling (liu2024review, 16). For instance, neural networks such as Cola-GNN (deng2020cola, 17) and DASTGN (pu2023dynamic, 18) have been developed for forecasting tasks, demonstrating superior performance compared to traditional mechanistic models. Additionally, hybrid models that integrate neural networks with mechanistic models, such as STAN  (gao2021stan, 19) and EINN  (rodriguez2023einns, 20), have been proposed to leverage existing epidemiological knowledge and enhance modeling accuracy.

Given the diversity of models available, there is a growing need for a unified platform to facilitate the study and development of computational methods for epidemic modeling. While existing packages in epidemic modeling (jenness2018epimodel, 21, 22, 23, 24, 25, 26, 27) offer tools like predefined models, simulation methods, and visualization aids, most have discontinued maintenance or failed to extend beyond traditional mechanistic models. In contrast, machine learning techniques have been thriving and demonstrating their versatility across domains  (xiao2023review, 28, 29). The trend in epidemic modeling is increasingly favoring neural and hybrid models  (liu2024review, 16, 30, 20). Therefore, there is an urgent need to develop a machine-learning-based epidemic library that encompasses a comprehensive set of cutting-edge models and fosters further research in this field.

In an effort to bridge the chasm between machine learning and epidemic modeling, this paper introduces EpiLearn, a Python-based library specifically designed for data mining within the realm of epidemiological data. The primary objective of our library is to offer a suite of user-friendly research tools that cater to both epidemiologists and data scientists alike. This endeavor is aimed at nurturing the evolution of innovative models and at enhancing the comprehension of the dynamics of epidemic spread. The core features of EpiLearn are delineated as follows:

  1. (a)

    Common Epidemiological Tasks. The library offers robust support for two prevalent tasks in epidemic modeling: forecasting and source detection. We have integrated automated pipelines to facilitate the training and evaluation of models within these domains.

  2. (b)

    Diverse Model Architectures. It encompasses a range of common model architectures, including Spatial, Temporal, and Spatial-Temporal baseline models, as well as exemplary epidemic models.

  3. (c)

    Data Simulation. The library is equipped with data simulation capabilities, such as the generation of random graphs and the simulation of epidemics on these graphs.

  4. (d)

    Transformations. A transformation module is provided to enable the processing and augmentation of epidemiological data, enhancing the flexibility and applicability of the models.

In summary, EpiLearn aims to streamline the research process by providing a unified framework for implementing, evaluating, and comparing various epidemic models. The library offers a modular and extensible architecture, allowing researchers to seamlessly incorporate new models, data sources, and evaluation metrics. It will facilitate collaboration and knowledge sharing among researchers, accelerate the development of new computational methods, and ultimately contribute to a better understanding and management of epidemic situations.

2. Related Work

The advancement of epidemic modeling and analysis has been greatly facilitated by the emergence of various software packages. Notable examples include EpiModel (jenness2018epimodel, 21), Epifit (doi2016epifit, 22), EpiEstim (coriepiestim, 23), Epinet (groendyke2018epinet, 24), EpiDynamics (santos2015epidynamics, 25), Eir (jacob2021eir, 26), and EoN (miller2020eon, 27). EpiModel, for instance, offers a versatile framework for modeling infectious diseases, employing deterministic, stochastic, and network-based approaches to enable a thorough analysis and simulation of epidemic dynamics.

However, these established packages encounter limitations when it comes to leveraging machine learning techniques, which represents a pivotal distinction between them and EpiLearn. In contrast to R-based packages such as EpiModel, EpiLearn is constructed on Python3 and makes extensive use of PyTorch during model development and assessment. This choice provides access to the extensive resources available within the Python and PyTorch ecosystems, preserving the capacity to integrate large pre-trained models and to perform efficient fine-tuning. Additionally, unlike Eir and EoN, EpiLearn transcends conventional mechanistic models by offering a diverse array of spatial, temporal, and spatial-temporal models. Moreover, EpiLearn encompasses a broader range of epidemic tasks, including source detection, and supplies streamlined pipelines for rapid model evaluation under these tasks. Finally, we introduce an interactive web application to enhance visualization capabilities, moving beyond the provision of static figures.

Refer to caption
Figure 1. Overview of EpiLearn.

3. Overview of EpiLearn

EpiLearn is an open-source Python library developed primarily using Python3 and PyTorch (paszkeautomatic, 31), along with other common toolkits such as PyG (fey2019fast, 32), NumPy (vanderwalt2011numpyb, 33), and Scikit-Learn (pedregosascikitlearn, 34). The aim of our package is to serve as a convenient tool for studying epidemic data and for building and evaluating epidemic models. In this section, we elaborate on the features provided by EpiLearn. The overall features are presented in Fig. 1.

3.1. Tasks in Epidemic Modeling

Following the taxonomy in  (liu2024review, 16), EpiLearn currently supports two tasks: Forecasting and Source Detection.

  1. (a)

    Forecasting. In the forecasting task, the model uses a given length of historical data, known as the lookback window, to predict multiple future steps, referred to as the horizon. For instance, the model predicts new infections over the next three days based on statistics from the past ten days. In this setting, both temporal and spatial-temporal models can be applied.

  2. (b)

    Source Detection. In the source detection task, the model aims to trace back the infection process and determine patient-zero in a contact graph. Some current researchers treat source detection as a classification task. The input is usually a graph at the current time, and the output is the probability of each node being the patient-zero.

In EpiLearn, each task is implemented as a Python class, which incorporates features such as model fitting, evaluation, and results. In Section 3.5, we further integrate each task into a pipeline.

3.2. Epidemic Models

EpiLearn supports multiple commonly used baselines as well as some epidemic-specific models. Based on the inputs, we categorize the models into spatial, temporal, and spatial-temporal models.

  1. (a)

    Temporal Models. Temporal models process only temporal data without additional spatial information. In EpiLearn, we include a wide range of types, such as an auto-regression model: ARIMA (shumway2017arima, 35), a machine learning model: XGBoost (chen2016xgboost, 36), mechanistic models: SIR, SIS, and SEIR (he2020seir, 8), and neural networks: GRU (cho2014learning, 37), LSTM (hochreiter1997long, 38), and DLinear (zeng2023transformers, 39). Additionally, we incorporate an epidemic-informed neural network model (rodriguez2023einns, 20).

  2. (b)

    Spatial Models. Spatial models utilize static features and static graphs as inputs. Common spatial baselines include GCN (kipf2016semi, 40), GAT (velickovic2017graph, 41), and GIN (xu2018powerful, 42). Currently, we use these models primarily for source detection tasks.

  3. (c)

    Spatial-Temporal Models. Spatial-temporal models incorporate both spatial and temporal information during inference. EpiLearn provides not only baselines but also several epidemic models developed in the past three years. For common baselines, we include CNNRNN (wu2018deep, 43), DCRNN (li2017diffusion, 44), ST-GCN (yu2017spatio, 45), and GraphWaveNet (wu2019graph, 46). For epidemic models, we include Cola-GNN (deng2020cola, 17), STAN (gao2021stan, 19), MepoGNN (cao2022mepognn, 47), EpiGNN (xie2022epignn, 48), EpiColaGNN (liu2023epidemiology, 49), and DASTGN (pu2023dynamic, 18). These models use dynamic features as input, although some can only process static graphs.

For each type of model, EpiLearn provides a unified framework for training and inference.

3.3. Data Processing

In addition to models, EpiLearn provides tools for pre-processing epidemic data. Currently, we have implemented normalization methods for both features and graph structures. Additionally, we offer useful transformations, such as converting features from the time domain to the frequency domain, adding time embeddings, and performing seasonal decomposition (cleveland1990stl, 50).

To ensure easy access to these data processing methods, EpiLearn follows a design similar to PyTorch Vision. During the pipeline construction, users can simply call the compose function and add the desired transformations as inputs.

3.4. Data Simulation

Similar to some R packages for epidemiology, EpiLearn provides simulation tools for quickly generating epidemic data. Specifically, we offer three types of simulations: spatial simulation, temporal simulation, and spatial-temporal simulation.

For spatial data, EpiLearn utilizes functions from PyG to generate random static graphs. When node features are available, node connections can be established by calculating the cosine similarity among nodes. For temporal data, EpiLearn uses mechanistic models introduced in Section 3.2 to simulate future health states over a given number of steps. For spatial-temporal data, EpiLearn employs TimeGEO (doi:10.1073/pnas.1524261113, 51) and NetworkSIR to simulate edge weights and node features across all regions over time.

These simulation tools enable users to create realistic epidemic scenarios for testing and validating models, facilitating research and application in epidemic modeling and analysis.

3.5. Epidemic Modeling Pipeline

To accelerate the training and evaluation of models on different datasets, EpiLearn provides a streamlined pipeline for various tasks, as illustrated in Fig. 1. At the beginning of the pipeline, we initialize the dataset with node features and states, static graphs, and dynamic graphs. Next, transformations are applied to the dataset.

For different tasks, we introduce a specific task class, such as forecasting, and input a prototype of the current model and dataset. Finally, we use the train function to train the current model on the dataset and the evaluate function to perform the evaluation.

3.6. Code Demonstration

To further show the convenience of EpiLearn, we provide a short demonstration below, which illustrates the building of the forecasting task. With a few lines of code listed below, we can easily train and evaluate a given model for forecasting tasks. This decoupling design of the task, dataset, and model also enables flexible changes made to the data or model, allowing swift evaluation on multiple models or datasets.

{python}

from epilearn.models.SpatialTemporal.STGCN import STGCN from epilearn.data import UniversalDataset from epilearn.utils import transforms from epilearn.tasks.forecast import Forecast # initialize settings lookback = 12 # inputs size horizon = 3 # predicts size # load toy dataset dataset = UniversalDataset() dataset.load_toy_dataset() # Adding Transformations transformation = transforms.Compose( ”features”:[transforms.normalize_feat()], ’graph’: [transforms.normalize_adj()]]) dataset.transforms = transformation # Initialize Task task = Forecast(STGCN, lookback, horizon) # Training result = task.train_model(dataset=dataset, loss=’mse’) # Evaluation evaluation = task.evaluate_model()

Refer to caption
Refer to caption
Figure 2. The interface developed for visualizing epidemic data.

3.7. Interactive Application

For further convenience, EpiLearn also builds a web application to visualize the epidemic data. At the current stage, we provide visualization of graph data and provide tools like tracing and showing variations of particular nodes over time. In addition, the web application also provides a simulation of the spread of the epidemic based on NetworkSIR. A demonstration is shown in Fig.  2.

4. Key Designs and Project Focus

In this section, we illustrate the key designs and focus of EpiLearn.

  1. (a)

    Modular and Decoupled Components. EpiLearn breaks down the training and evaluation process into distinct components: model, dataset, transformations, and configurations. For every model, we provide unified training and prediction functions. The dataset component, which offers multiple features such as data splitting, requires a few key inputs, including node features and graphs. Regarding transformations, we adopt an extensible design similar to the module in PyTorch, where transformation functions are aggregated in the Compose function.

  2. (b)

    User-Friendly Framework. Building on the aforementioned modules, we offer a unified framework for training and evaluating models and datasets on specific tasks. Users can easily switch among different models and datasets within our designed pipelines and modify configurations as needed. For epidemiologists and data scientists, EpiLearn also provides an interface to present data overviews and other useful tools like simulation.

  3. (c)

    Extensibility. The modular design of our library also ensures extensibility. Researchers interested in testing or developing new epidemic models can follow a structure similar to the baseline models. By creating a new file for the PyTorch model or a new function for transformations, users can easily implement and test their methods.

  4. (d)

    Reproducibility. To ensure reproducibility, we provide detailed documentation on setting up the environment, including the required dependencies and versions. We also include scripts for downloading datasets and running experiments with fixed random seeds to minimize result variations. The code and documentation for this project are publicly available at: https://github.com/Emory-Melody/EpiLearn.

5. Impacts and Future Plans

Impacts. This paper presents EpiLearn: a Python toolkit for epidemic modeling and analysis. We provide various features, including task pipelines, data simulation, and transformations, aiming for EpiLearn to have a positive impact on both the machine learning and epidemiology domains.

  1. (a)

    Data Scientists’ Perspective. For data scientists, EpiLearn serves as a comprehensive toolbox and framework for quickly building and evaluating epidemic models. The modular design allows users to easily modify and add new features and models. Additionally, the unified pipeline facilitates fast training and testing of models on different tasks, simplifying the process of benchmarking existing models.

  2. (b)

    Epidemiologists’ Perspective. For epidemiologists, EpiLearn provides several epidemic models and tasks, along with simulation methods and interactive applications for visualizing epidemic data. These tools assist users in understanding the dynamics of complex systems, testing empirical hypotheses about outbreak trajectories, and gaining an overview of interventions such as vaccinations and quarantines.

Future Plans. Looking ahead, we plan to expand EpiLearn to include more general epidemic tasks such as projection and surveillance, along with related baselines and epidemic models. Continuous updates to data processing tools will enhance model training and data analysis. Furthermore, additional simulation functions will be developed to aid decision-making during epidemic crises. Lastly, ongoing development of our web application aims to provide a more comprehensive and user-friendly analysis experience.

References

  • (1) Philip E Paré, Carolyn L Beck and Tamer Başar “Modeling, estimation, and analysis of epidemics over networks: An overview” In Annual Reviews in Control 50 Elsevier, 2020, pp. 345–360
  • (2) Yue Xiang et al. “COVID-19 epidemic prediction and the impact of public health interventions: A review of COVID-19 epidemic models” In Infectious Disease Modelling 6 Elsevier, 2021, pp. 324–342
  • (3) Shun Zheng et al. “Hierst: A unified hierarchical spatial-temporal framework for covid-19 trend forecasting” In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, 2021, pp. 4383–4392
  • (4) Dennis L Chao, M Elizabeth Halloran, Valerie J Obenchain and Ira M Longini Jr “FluTE, a publicly available stochastic influenza epidemic simulation model” In PLoS computational biology 6.1 Public Library of Science San Francisco, USA, 2010, pp. e1000656
  • (5) Hao Sha, Mohammad Al Hasan and George Mohler “Source detection on networks using spatial temporal graph convolutional networks” In 2021 IEEE 8th International Conference on Data Science and Advanced Analytics (DSAA), 2021, pp. 1–11 IEEE
  • (6) Sirui Song et al. “Reinforced epidemic control: Saving both lives and economy” In arXiv preprint arXiv:2008.01257, 2020
  • (7) B Jhun “Effective vaccination strategy using graph neural network ansatz (2021)”
  • (8) Shaobo He, Yuexi Peng and Kehui Sun “SEIR modeling of the COVID-19 and its dynamics” In Nonlinear dynamics 101 Springer, 2020, pp. 1667–1680
  • (9) Kaiming He, Xiangyu Zhang, Shaoqing Ren and Jian Sun “Deep Residual Learning for Image Recognition” In CVPR, 2016, pp. 770–778
  • (10) Saining Xie et al. “Aggregated residual transformations for deep neural networks” In CVPR, 2017, pp. 1492–1500
  • (11) Andrew G Howard et al. “Mobilenets: Efficient convolutional neural networks for mobile vision applications” In arXiv preprint arXiv:1704.04861, 2017
  • (12) Ashish Vaswani et al. “Attention is all you need” In NeurIPS 30, 2017
  • (13) Thomas N Kipf and Max Welling “Semi-supervised classification with graph convolutional networks” In ICLR, 2017
  • (14) Petar Veličković et al. “Graph attention networks” In arXiv preprint arXiv:1710.10903, 2017
  • (15) Will Hamilton, Zhitao Ying and Jure Leskovec “Inductive representation learning on large graphs” In NeurIPS, 2017
  • (16) Zewen Liu et al. “A review of graph neural networks in epidemic modeling” In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2024
  • (17) Songgaojun Deng et al. “Cola-gnn: Cross-location attention based graph neural networks for long-term ili prediction” In Proceedings of the 29th ACM international conference on information & knowledge management, 2020, pp. 245–254
  • (18) Xiaojun Pu et al. “Dynamic adaptive spatio–temporal graph network for COVID-19 forecasting” In CAAI Transactions on Intelligence Technology Wiley Online Library, 2023
  • (19) Junyi Gao et al. “STAN: spatio-temporal attention network for pandemic prediction using real-world evidence” In Journal of the American Medical Informatics Association 28.4 Oxford University Press, 2021, pp. 733–743
  • (20) Alexander Rodríguez et al. “Einns: Epidemiologically-informed neural networks” In Proceedings of the AAAI Conference on Artificial Intelligence 37.12, 2023, pp. 14453–14460
  • (21) Samuel M Jenness, Steven M Goodreau and Martina Morris “EpiModel: an R package for mathematical modeling of infectious disease over networks” In Journal of statistical software 84 NIH Public Access, 2018
  • (22) K Doi, K Sakabe and M Taruri “epifit: Flexible Modelling Functions for Epidemiological Data Analysis” In R package version 0.0 6, 2016
  • (23) Anne Cori et al. “Package ‘EpiEstim”’ In CRAN: Vienna Austria 13, 2020
  • (24) Chris Groendyke and David Welch “epinet: an R package to analyze epidemics spread across contact networks” In Journal of Statistical Software 83, 2018, pp. 1–22
  • (25) O Santos Baquero and F Silveira Marques “Epidynamics: dynamic models in epidemiology” In R package version 0.3. 0, URL https://CRAN. R-project. org/package= EpiDynamics, 2015
  • (26) Mathew Jacob “Eir: A Python Package for Epidemic Simulation” In Journal of Open Source Software 6.62, 2021, pp. 3247
  • (27) Joel C Miller and Tony Ting “Eon (epidemics on networks): a fast, flexible python package for simulation, analytic approximation, and analysis of epidemics on networks” In arXiv preprint arXiv:2001.02436, 2020
  • (28) Dannier Xiao, Mehrdad Dianati, William Gonçalves Geiger and Roger Woodman “Review of graph-based hazardous event detection methods for autonomous driving systems” In IEEE Transactions on Intelligent Transportation Systems 24.5 IEEE, 2023, pp. 4697–4715
  • (29) David Ahmedt-Aristizabal et al. “Graph-based deep learning for medical diagnosis and analysis: past, present and future” In Sensors 21.14 MDPI, 2021, pp. 4758
  • (30) Wyatt Madden et al. “Neural networks for endemic measles dynamics: comparative analysis and integration with mechanistic models” In medRxiv Cold Spring Harbor Laboratory Press, 2024, pp. 2024–05
  • (31) Adam Paszke et al. “Automatic Differentiation in PyTorch”
  • (32) Matthias Fey and Jan Eric Lenssen “Fast Graph Representation Learning with PyTorch Geometric” arXiv, 2019 eprint: 1903.02428
  • (33) Stéfan Van Der Walt, S Chris Colbert and Gaël Varoquaux “The NumPy Array: A Structure for Efficient Numerical Computation” In Computing in Science & Engineering 13.2, 2011, pp. 22–30
  • (34) Fabian Pedregosa et al. “Scikit-Learn: Machine Learning in Python” In MACHINE LEARNING IN PYTHON
  • (35) Robert H Shumway, David S Stoffer, Robert H Shumway and David S Stoffer “ARIMA models” In Time series analysis and its applications: with R examples Springer, 2017, pp. 75–163
  • (36) Tianqi Chen and Carlos Guestrin “Xgboost: A scalable tree boosting system” In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, 2016, pp. 785–794
  • (37) Kyunghyun Cho et al. “Learning phrase representations using RNN encoder-decoder for statistical machine translation” In arXiv preprint arXiv:1406.1078, 2014
  • (38) Sepp Hochreiter and Jürgen Schmidhuber “Long short-term memory” In Neural computation 9.8 MIT press, 1997, pp. 1735–1780
  • (39) Ailing Zeng, Muxi Chen, Lei Zhang and Qiang Xu “Are transformers effective for time series forecasting?” In Proceedings of the AAAI conference on artificial intelligence 37.9, 2023, pp. 11121–11128
  • (40) Thomas N Kipf and Max Welling “Semi-supervised classification with graph convolutional networks” In arXiv preprint arXiv:1609.02907, 2016
  • (41) Petar Velickovic et al. “Graph attention networks” In stat 1050.20, 2017, pp. 10–48550
  • (42) Keyulu Xu, Weihua Hu, Jure Leskovec and Stefanie Jegelka “How powerful are graph neural networks?” In arXiv preprint arXiv:1810.00826, 2018
  • (43) Yuexin Wu, Yiming Yang, Hiroshi Nishiura and Masaya Saitoh “Deep learning for epidemiological predictions” In The 41st international ACM SIGIR conference on research & development in information retrieval, 2018, pp. 1085–1088
  • (44) Yaguang Li, Rose Yu, Cyrus Shahabi and Yan Liu “Diffusion convolutional recurrent neural network: Data-driven traffic forecasting” In arXiv preprint arXiv:1707.01926, 2017
  • (45) Bing Yu, Haoteng Yin and Zhanxing Zhu “Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting” In arXiv preprint arXiv:1709.04875, 2017
  • (46) Zonghan Wu et al. “Graph wavenet for deep spatial-temporal graph modeling” In arXiv preprint arXiv:1906.00121, 2019
  • (47) Qi Cao et al. “Mepognn: Metapopulation epidemic forecasting with graph neural networks” In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, 2022, pp. 453–468 Springer
  • (48) Feng Xie et al. “EpiGNN: Exploring spatial transmission with graph neural network for regional epidemic forecasting” In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, 2022, pp. 469–485 Springer
  • (49) Mutong Liu, Yang Liu and Jiming Liu “Epidemiology-aware Deep Learning for Infectious Disease Dynamics Prediction” In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, 2023, pp. 4084–4088
  • (50) Robert B Cleveland, William S Cleveland, Jean E McRae and Irma Terpenning “STL: A seasonal-trend decomposition” In J. Off. Stat 6.1, 1990, pp. 3–73
  • (51) Shan Jiang et al. “The TimeGeo modeling framework for urban mobility without travel surveys” In Proceedings of the National Academy of Sciences 113.37, 2016, pp. E5370–E5378 DOI: 10.1073/pnas.1524261113