1 A fairness assessment of mobility -based COVID -19 case prediction models Abdolmajid Erfani 1 and Vanessa Frias -Martinez 2 3

2025-04-27 0 0 1.1MB 25 页 10玖币

侵权投诉

A fairness assessment of mobility-based COVID-19 case prediction models

Abdolmajid Erfani 1*, and Vanessa Frias-Martinez 2, 3

1 Department of Civil and Environmental Engineering, University of Maryland, 1173 Glenn

Martin Hall, College Park, MD 20742, USA.

2 College of Information Studies, University of Maryland, College Park, MD 20742, USA.

3 University of Maryland Institute for Advanced Computer Studies, University of Maryland,

College Park, MD 20742, USA

* Corresponding Author: erfani@umd.edu

ABSTRACT

In light of the outbreak of COVID-19, analyzing and measuring human mobility has become

increasingly important. A wide range of studies have explored spatiotemporal trends over time,

examined associations with other variables, evaluated non-pharmacologic interventions (NPIs),

and predicted or simulated COVID-19 spread using mobility data. Despite the benefits of publicly

available mobility data, a key question remains unanswered: are models using mobility data

performing equitably across demographic groups? We hypothesize that bias in the mobility data

used to train the predictive models might lead to unfairly less accurate predictions for certain

demographic groups. To test our hypothesis, we applied two mobility-based COVID infection

prediction models at the county level in the United States using SafeGraph data, and correlated

model performance with sociodemographic traits. Findings revealed that there is a systematic bias

in models’ performance toward certain demographic characteristics. Specifically, the models tend

to favor large, highly educated, wealthy, young, and urban counties. We hypothesize that the

mobility data currently used by many predictive models tends to capture less information about

older, poorer, less educated and people from rural regions, which in turn negatively impacts the

accuracy of the COVID-19 prediction in these areas. Ultimately, this study points to the need of

improved data collection and sampling approaches that allow for an accurate representation of the

mobility patterns across demographic groups.

INTRODUCTION

The interactions between human mobility and epidemic spread have been studied unprecedentedly

during the COVID-19 pandemic 1-8. With these efforts, nonpharmaceutical interventions (such as

national lockdowns) have been evaluated for their effectiveness and socio-economic impact on

different groups 9-11, models have been developed to predict disease spatial diffusion 12,13, and

scenarios have been modeled to assess their outcomes 14-17. Studies have demonstrated that

mobility data are a meaningful proxy measure of social distancing 18, affect viral spreading 19,20,

and are useful for predicting the spread of COVID-19 21-23.

In particular, to control the spread of new cases and plan efficiently for hospital needs and

capacities during an epidemic, public health decision-makers require accurate predictions of future

case numbers 7. For example, a study by Ilin et al. (2021) showed that changes in mobility can be

used to predict COVID-19 cases. Their study demonstrated that public mobility data can be used

to develop reduced-form and simple models that mimic the behavior of more sophisticated

epidemiological models for predicting COVID-19 cases on a 10-day basis 21. Another study

examined several state-of-the-art machine learning models and statistical methods and

demonstrated how mobility data can improve prediction trends when used as exogenous

information in models 22.

As discussed, mobility data from anonymized smartphones has been shown to improve COVID-

19 case prediction models. However, mobility data bias has received little attention in this

predictive context. There exist only just a handful of papers reporting demographic bias in mobility

data due to differences in smartphone ownership and use 24-26; and since data providers are not

transparent about how mobility data is collected, or about the socio-economic and demographic

groups represented in them, directly measuring and correcting bias in mobility data is difficult. In

this study, we hypothesize that the presence of socio-economic and demographic bias in the

mobility data used to train the COVID-19 case predictive models, might result in unfairly less

accurate predictions for particular socio-economic and demographic groups. Unfair predictions

provided to decision makers e.g., predictions of COVID-19 cases for minority groups that are

lower than reality, could in turn be used to unfairly assign more resources to population groups

that do not necessarily need them.

To test our hypothesis, we evaluated the performance of two types of mobility-based COVID-19

case prediction models highly used by decision makers due to its interpretability: linear regressions

and time series models. In contrast to more complex epidemiological models that are hard to tune

due to its parametric nature, and deep learning models that are difficult to interpret, linear models

and time series are easy to train and test 21,27-29. The models were trained using SafeGraph’s

mobility data, and performance was measured via predictive errors. To assess the fairness of the

predictions, we analyzed the relationship between the model prediction errors and specific socio-

economic and demographic features at the county level in the United States and across the two

model types. Evaluating the performance of two diverse interpretable models allowed us to

account for potential algorithmic bias i.e., bias introduced by the algorithm itself 30,31. If unfair

predictions are pervasive across types of models trained and tested with the same data, we can

partially attribute the unfairness to the mobility data itself.

MATERIAL AND METHODS

In our study, we use mobility data from SafeGraph to build COVID-19 case prediction models;

and we explore model performance across socio-economic and demographic features to potentially

identify unfair results for specific groups i.e., differences in error distributions across social groups.

We next describe these three types of datasets, with all being publicly available.

Human mobility

We used SafeGraph’s publicly-available human mobility data at the county level in the US.

SafeGraph uses location information extracted from smartphones to provide aggregate data

characterizing mobility in terms of visit volumes to types of places and volumes of origin-

destination (OD) flows 32. For this study specifically, we used the data publicly available in the

origin-destination-time (ODT) platform 33, that computes OD flows between counties as the

aggregation of trips that start at an individual’s home county location (origin), with a destination

defined as a stay location within a county for longer than a minute. OD flows between all counties

in the US were collected throughout all days of the year 2020. Figure 1.a illustrates how the

average number of trips at county level across the US changed over the year 2020. According to

various studies in the US using mobility data, the dataset collected in Figure 1.a also shows similar

trends of mobility change 34,35.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

1Afairnessassessmentofmobility-basedCOVID-19casepredictionmodelsAbdolmajidErfani1*,andVanessaFrias-Martinez2,31DepartmentofCivilandEnvironmentalEngineering,UniversityofMaryland,1173GlennMartinHall,CollegePark,MD20742,USA.2CollegeofInformationStudies,UniversityofMaryland,CollegePark,MD20742,USA.3Univ...

展开>> 收起<<

1 A fairness assessment of mobility -based COVID -19 case prediction models Abdolmajid Erfani 1 and Vanessa Frias -Martinez 2 3.pdf

共25页,预览5页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

1 A fairness assessment of mobility -based COVID -19 case prediction models Abdolmajid Erfani 1 and Vanessa Frias -Martinez 2 3

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: