Modeling impact of Covid-19 pandemic on spatial temporal human mobility

•A novel computational and time efficient methodology for forecasting human mobility under conditions of NPIs imposition is developed.•The methodology combines data quality improvement and dimensionality reduction techniques with a DNN algorithm.•Training employs grid search hyperparameter tuning an...

Full description

Saved in:

Bibliographic Details
Published in:	Expert systems with applications Vol. 297; p. 128972
Main Author:	Titus-Glover, Leslie
Format:	Journal Article
Language:	English
Published:	Elsevier Ltd 01.02.2026
Subjects:	Artificial intelligence Connection weight relative attribute importance method Deep neural networks Epidemiological surveillance, resilience, vulnerability, and non-pharmacological intervention attributes Pearson and Cramér’s V correlation analysis Spatial temporal human mobility ANN CNN DNN NPI GLM KNN GPS SVM Spatial temporal human mobility SVR MLP DBN RNN IHME Connection weight relative attribute importance method SEIR Epidemiological surveillance, resilience, vulnerability, and non-pharmacological intervention attributes ML WHO R2 Deep neural networks LSTM Covid-19 AAE OxCGRT AI UMA Pearson and Cramér’s V correlation analysis MLPNN RF ARIMA U.S FAST Artificial intelligence MRRG
ISSN:	0957-4174
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	•A novel computational and time efficient methodology for forecasting human mobility under conditions of NPIs imposition is developed.•The methodology combines data quality improvement and dimensionality reduction techniques with a DNN algorithm.•Training employs grid search hyperparameter tuning and connection weight input pruning methods to produce an optimized DNN model.•The trained DNN model, tested on simulated counterfactual NPI scenarios, reveals findings that compares well with real-world expectations.•Novel methodology demonstrates high accuracy and generalization, and efficient resource utilization. Physics-derived and statistical algorithms are utilized for modeling spatial–temporal human mobility. Post Covid-19, use of SEIR and GLM frameworks incorporating epidemiological, NPIs, and human mobility attributes have been investigated to further understanding of NPI impact on human mobility. All of these approaches have reported limited effectiveness due to various constraints including theoretical models holding true to physical boundaries, statistical models adhering to Gaussian and other assumptions, complexity and resource demand of SEIR models, and simplicity of GLM for modeling human mobility patterns. This paper introduces a novel computational and time efficient methodology for forecasting human mobility under conditions of NPI imposition. Methodology integrates statistical data quality improvement and dimension-reduction techniques with DNN algorithms. It is optimized via grid search hyperparameter tuning and pruning input attributes based on weighted connectivity contributions. Training on 11,418 records produced DNN model optimal configuration comprising 19 inputs including 6 NPIs, 3 hidden layers with 8 neurons each, and output—predicted change in human mobility. Training goodness of fit statistics were: R2 = 0.9423 and average absolute error (AAE) = 0.03282 while validation on 2,898 set aside records yielded R2 of 0.9432 and AAE of 0.03147. Counterfactual scenarios simulating NPI impact on human mobility reveals workplace closures; restrictions on gatherings and public transport; and stay-at-home orders reduced human mobility. The methodology, demonstrates high prediction accuracy, ability to generalize, and efficient resource utilization processing unstructured, high-dimensional data and forecasting human mobility. It fills the gap when resource-efficient models with high predictive accuracy is lacking.
ISSN:	0957-4174
DOI:	10.1016/j.eswa.2025.128972