Attention LSTM과 SHAP을 사용한 설명 가능한 COVID-19 확진자 수 예측 기법 = Explainable COVID-19 Forecasting Scheme Using Attention LSTM and SHAP

다국어 초록 (Multilingual Abstract)

Due to the spread of COVID-19, many losses are being reported worldwide. In order to establish effective countermeasures to prevent the spread of COVID-19, it is necessary to accurately predict the extent of the spread of COVID-19. However, due to the recent introduction of vaccines, a forecasting model suitable for the changed diffusion pattern is needed, and although the forecasting accuracy has been improved through a machine learning-based approach, the model does not provide sufficient confidence due to the lack of explanatory properties of the model. Therefore, in this paper, we propose an explainable forecasting scheme that forecasts the number of COVID-19 confirmed cases using an Attention LSTM(Long Short-Term Memory) model and explains the analysis results using SHAP(SHapley Additive exPlanations). Changes due to introduction of vaccines were reflected by collecting various data related to COVID-19, including vaccination data, and using it as an input variable for the Attention LSTM model. Through comparative experiments with various models, we demonstrated the excellent forecasting performance of the proposed model and the explanatory capacity of the SHAP for the results.

국문 초록 (Abstract)

COVID-19의 확산으로 인해 세계적으로 많은 손실이 보고되고 있다. 효과적인 확산 방지 대책을 수립하기 위해서는 COVID-19 확산 정도의 정확한 예측이 필요하며, 이를 위해 확산 초기부터 기계적 접근법이나 기계학습 기반 접근법 등을 활용한 기법들이 제안되어왔다. 그러나 최근 백신 도입으로 인하여 변화된 확산 패턴에 적합한 예측 모델이 필요하며, 기계학습 기반 접근법을 통해 예측 정확도는 높아졌으나 모델의 설명성이 부족하여 충분한 신뢰를 주지 못하는 실정이다. 이에, 본 논문에서는 Attention LSTM(Long Short-Term Memory) 모델을 사용하여 COVID-19 확진자 수를 예측하고, 그 결과를 SHAP(SHapley Additive exPlanations)을 통하여 분석하는 설명 가능한 COVID-19 확진자 수 예측 기법을 제안한다. 백신 접종 데이터를 포함한 COVID-19 관련 다양한 데이터를 수집하고, Attention LSTM 모델의 입력 변수로 사용함으로써 백신 도입으로 인한 변화를 반영하였다. 다양한 모델과의 비교 실험을 통하여 제안한 모델의 우수한 예측 성능을 보였으며, SHAP을 통해 예측 결과에 대한 설명 가능성을 입증하였다.

참고문헌 (Reference)

1 "Tensorflow"

2 Liao, Zhifang, "TW-SIR: time-window based SIR for COVID-19 forecasts" 10 (10): 1-15, 2020

3 Alene, Muluneh, "Serial interval and incubation period of COVID-19: a systematic review and meta-analysis" 21 (21): 1-9, 2021

4 "Scikit-learn"

5 Bastos, Saulo B., "Modeling and forecasting the Covid-19 pandemic in Brazil"

6 Sak, Hasim, "Long short-term memory recurrent neural network architectures for large scale acoustic modeling" 2014

7 Jin, Xiaoyong, "Inter-Series Attention Model for COVID-19Forecasting" Society for Industrial and Applied Mathematics 2021

8 Yang, Zichao, "Hierarchical attention networks for document classification" 2016

9 Kumar, Pavan, "Forecasting the dynamics of COVID-19 pandemic in top 15 countries in April 2020: ARIMA model with machine learning approach"

10 Prasanth, Sikakollu, "Forecasting spread of COVID-19 using Google Trends: A hybrid GWO-Deep learning approach" 142 : 110336-, 2021

11 ArunKumar, K. E., "Forecasting of covid-19using deep layer recurrent neural networks (rnns)with gated recurrent units (grus) and long short-term memory (lstm) cells" 146 : 110861-, 2021

12 D. Gunning, "Explainable artificial intelligence (XAI)" Defense Advanced Research Projects Agency (DARPA) 2017

13 Atkeson, Andrew, "Estimating and forecasting disease scenarios for COVID-19 with an SIR model" National Bureau of Economic Research 2020

14 Fernandes, Nuno, "Economic effects of coronavirus outbreak (COVID-19) on the world economy" SSRN 2020

15 Zhang, Nan, "Dynamic population flow based risk analysis of infectious disease propagation in a metropolis" 94 : 369-379, 2016

16 Rodriguez, Alexander, "Deepcovid: An operational deep learning-driven framework for explainable real-time covid-19 forecasting" 2020

17 Chandra, Rohitash, "Deep learning via LSTM models for COVID-19 infection forecasting in India"

18 Sallam, Malik, "COVID-19 vaccine hesitancy worldwide: a concise systematic review of vaccine acceptance rates" 9 (9): 160-, 2021

19 Rustam, Furqan, "COVID-19 future forecasting using supervised machine learning models" 8 : 101489-101499, 2020

20 Ghostine, Rabih, "An extended seir model with vaccination for forecasting the covid-19 pandemic in saudi arabia using an ensemble kalman filter" 9 (9): 636-, 2021

21 Kingma, Diederik P., "Adam: A method for stochastic optimization"

22 Lundberg, Scott M., "A unified approach to interpreting model predictions" 2017

23 Mathieu, Edouard, "A global database of COVID-19 vaccinations" 1-7, 2021