Skip Navigation

jinfang wANG, PHD

Yokohama City University
Professor and Dean of the School of Data Science

Jinfang Wang was appointed Dean of School of Data Science, and Chairperson of Division of Data Science, Graduate School of Data Science, Yokohama City University, Japan in April 2020.

Before joining Yokohama City University, he was Professor, Department of Mathematics, Chiba University, Japan, teaching and doing research in Statistics. Prior to Chiba, Dr. Wang was adjunct Associate Professor, Open University, Japan; Associate Professor, Department of Animal Husbandry, Obihiro University of Agriculture and Veterinary Medicine, Japan; Visiting Assistant Professor, Department of Statistics and Actuarial Science, University of Waterloo, Canada; Assistant Professor, the Institute of Statistical Mathematics, Japan.

Dr. Wang holds a bachelor’s in mathematics from Yangzhou University, China, a master’s and PhD degree in Statistics from Chiba University Japan. Dr. Wang started his career in 1996 as a researcher in Statistics at the Institute of Statistical Mathematics, Japan, a top research institute in Statistics in Asia and the world as well. He has published many books and research papers and given many invited lectures. He has also served many professional committees, including Director of the Japan Statistical Society, the Vice Chairperson of School of Statistical Thinking, the Institute of Statistical Mathematics, the Chairperson of Higher Certificate and Graduate Diploma jointly sponsored by the Royal Statistical Society & the Japan Statistical Society.


A Convolutional LSTM Model for Predicting the SARS-COV-2 Positives in Japan

The number of newly infected people with COVID-19 is still increasing at the worldwide level. So far, a large number of papers have been published on predicting the number of infected people. Nevertheless, most of these papers are based on the SIR model and its variants, the basic infectious disease dynamics models developed in the early 1900s. SIR models require strong unrealistic assumptions for modeling real disease dynamics, which is particularly so for the current COVID-19 pandemic. To overcome these difficulties, we propose a deep learning-based method. The proposed method is a neural network model consisting of appropriately designed convolutional LSTM layers, which enable incorporation of relevant temporal and geographic covariates, including the daily number of COVID-19 positives.

We will demonstrate the usefulness of the proposed method by applying it to the temporal-spatial data of the 23 wards of Tokyo to predict the daily number of new positives of COVID-19 in short-term prediction approach. We use the data from April 2, 2020 to November 30, 2020 for training the model, and data from December 1, 2020 to December 31, 2020 for testing the performance of the model. Since convolutional LSTM layers require that the inputs are images, we mapped the numbers of daily new positives and the covariates in corresponding wards to the map of Tokyo. Then we make windows of consecutive samples from the series of static images we made. The model makes a next day prediction based on one of the windows of it. We iterate prediction for the testing period. Our experiments demonstrated that our method outperforms SIR models in terms of MSE and MAE.