Evaluation of COVID-19 Pandemic in Six Asia-Pacic Countries using Data-Driven and Machine Learning Method Based on the Current Management and Interventions

Background South-east Asia and Western Pacic countries have large populations and underreporting of Covid19, which pose challenges to the large-scale response. Methods Data-driven methods are used to evaluate the Government or society’s interventions and the situation of the COVID-19 pandemic, and machine learning method are used to forecast the trend of COVID-19 pandemic based on the current management and interventions. Results The results show that. India received low government response index scores in February, and the number of conrmed cases and active cases in September became quite high with large stock and the overall growth rate is higher than 1. The number of daily conrmed cases in Bangladesh, Japan and Philippines is low and on the decline, it is rising in Malaysia and Indonesia. The number of active cases in Bangladesh, Japan, India and Bangladesh has begun to decline, Malaysia and Indonesia is no sign of decline. Bangladesh, Japan and Philippines will be at or moderating, while Malaysia and Indonesia will still have no slowdown momentum and the situation will be severe. Conclusions The results show that the existing management and interventions responses are effective, although they have room for improvement, and Malaysia and Indonesia need to be improved and strengthened.


Introduction
Real-time data show that the con rmed cases of COVID-19 are growing like crazy in most countries of the world [1] [2]. Countries or regions of the world have adopted different social, economic, health and environmental measures to intervene in the epidemic, there is with no vaccine or drug are available, which interventions focus on social distancing and movement restrictions, contact tracing, isolation of suspected and con rmed cases, Lockdowns and public health measures [3] [4,5]. However, the COVID-19 transmission continues to spread in many countries or regions and has not achieve control. It is still a major public health crisis and in the world although some countries, like South Korea and China, are gradually lifting these measures after achieving effective control over transmission [1] [6] [3]. However, long or intermittent governmental interventions and responses may be required to reduce the possibility of infection recurrence.
The countries of South-east Asia and Western Paci c are among the rst to experience COVID-19 outbreaks with huge populations and relatively scarce public health resources, so the risk of pandemics of infectious diseases is relatively high. At the same time, high levels misdiagnosis and of under-reporting make a challenges for government to large-scale responses [7] [8]. Therefore, it is necessary to conduct correct evaluation, modeling and forecasting studies on epidemic data to estimate the potential impact of interventions during the uncertain phase of the epidemic and to provide reference and guidance for the next step of epidemic prevention and control, which is critical if local transmission continues to increase [9] [10] [11] [5].
Driven by a large amount of data from our life and health record, through interdisciplinary exploration in data science, statistics, network science, physics and epidemiology, a simple mathematical models describing the essence of the spread of epidemics can be generated and used to simulate and t data with a number of appropriate parameters and make informed forecast [12] [13] [3] [14]. A large amount of evidence in recent decades shows that mathematical modeling and prediction have made great progress in better understanding the spread and transmission dynamics of infectious diseases, and played a key role in nding the optimal containment strategies in the spread of epidemics [15] [16] [17] [18] [19].
In this study, under the existing interventions and responses, we conducted data-driven comprehensive analysis and machine learning method to evaluate and forecast the COVID-19 transmission in countries with top three con rmed cases in Western Paci c and South-east Asia as of Jan 30, 2021. It aims to assess the interventions and responses in these countries and the status of the pandemic, establish a national spatial model of the spread of COVID-19 to estimate the time distribution and development trend of COVID-19 in each country. We would be pleased if our work could contribute to further research and applications that could help the effective interventions of the COVID-19 epidemic.

Data Source
The three countries with the largest number of cumulative con rmed COVID-19 cases in South-east Asia and western paci c on Jan 30, 2021 were selected as study subjects. Data Source: 2019 Novel

Governmental interventions and responses
The interventions and responses were analyzed as response stringency Index through R "tidycovid19" package [25]. The OxCGRT project composite measure of nine of metrics (stay-at-home requirements, restrictions on public gatherings, school closures, public information campaigns, workplace closures, closures of public transport, cancellation of public events, international travel controls and restrictions on internal movements) to calculate a government stringency Index. The index score of the nine metrics between 0 and 100 on any given day.

Assessment of COVID-19 pandemic based on the current interventions
The overall situation, daily changes and growth rates, total con rmed cases per million people, total deaths per million people and case fatality rate (CFR) of the ongoing COVID-19 pandemic based on the current interventions and responses in each country were compute using "EpiModel" package and "COVID19.Analytics" package in R [26].

The trend of COVID-19 pandemic based on the current policy environment
Machine Learning method[27] using to forecasting the trend of COVID-19 pandemic based on the existing interventions and responses in Japan, Philippines, Malaysia, India, Bangladesh and Indonesia and Generating a 365 day ahead forecast of con rmed cases of COVID-19 with 95% prediction interval (PI).
The algorithm and formula are: A basic model was established with no tweaking of season-related parameters and additional regression elements. Machine Learning method were used to model and predict the number of active cases in each country. The Machine Learning method will assign each day in future a predicted value which it names yhat, Yhat_lower and Yhat_upper are the upper and lower yhat of the prediction (based on 95% prediction Interval).

Results
As of Jan 30, 2021, the three countries with the largest number of cumulative con rmed COVID-19 cases in South-east Asia and western paci c include Japan, the Philippines, Malaysia, India, Bangladesh and Indonesia, which we selected as study subjects.

Government interventions and responses
The interventions and responses were evaluated as Response Stringency Index through R package "Tidycovid19" to summarize the severity of the interventions and responses, a higher index score indicates a government response stricter. The median interventions and responses index is 38.89 for Japan, 80.09 for Bangladesh, 73.61 for India, 68.98 for the Philippines, 59.72 for Indonesia and 57.41 for Malaysia (Figure 1 a, b) . Total con rmed cases per million people of the ongoing COVID-19 pandemic is 3084.89 (Japan), 3249.38 (Bangladesh), 7795.35 (India), 4796.61 (the Philippines), 3942.31 (Indonesia) and 6641.51 (Malaysia) (Figure 1 b Figure S1). In general, the number of con rmed cases per million decreased with the increase of the government responses stringency index (Figure 1. a, b). India received low government response index scores in February, and as a result, the number of cases was already very high in September. It can be seen that the government's early anti-epidemic measures were very important.

The overall situation of the epidemic
The total number of con rmed cases, death, recovered and active cases show in Figure2 in Jan 30, 2021. Japan had 387499 con rmed COVID-19 cases, 325436 recovered, 5688 deaths and 56375 active cases, indicating a downward trend in the number of active cases (Figure 2a). In the Philippines, 523516 con rmed cases, 475904 recovered, 10669 deaths and 36943 active cases have been con rmed with a wave (Figure 2b). The number of con rmed COVID-19 cases was 209661, recovered patients was 161527, death was 746, and the number of active cases was 47388 with a going up in waves in Malaysia (Figure 2c). In India, there were 10.74617 M con rmed cases, 10.42311 M recovered, 154274 death and 168791active cases, indicating that the number of active cases showed a decreasing trend (Figure 2d). 534770 con rmed cases, 479297 recovered, 8111deaths and 47362 active cases have been reported In Bangladesh, with the number of active cases showed a decreasing trend (Figure 2e). Indonesia had 1.066313 M con rmed cases, 862502 recovered, 29728 deaths, and 174083 active cases. The number of active cases and con rmed cases is still on the rise in Indonesia (Figure 2f).
The Figure 3 show the total number of cases for six countries with a con dence band based on moving average and overall growth rate. Plot the number of cases as a function of time for the given locations and type of categories to generate different ts to match the data in a log-scale scatter and a linear scale bar plot. The overall growth rate is closer to 1 in Japan and Malaysia (Figure 3a and c), indicating its almost stopped growth, while in Philippines and Indonesia is always higher than 1 (Figure 3b and f), the number of cases in these countries is constantly increasing, especially in India and Bangladesh (lm-exp GR = 1.04), where the growth rate is much higher than 1 (Figure 3d and e), indicating its close to the exponential growth.

Daily changes and growth rates
Daily con rmed cases changes were computed that display two scatter plots per country with the number of changes, both in log-scale (right vertical axis) and linear scale (left vertical axis) combined in upper panel. The growth rate between two consecutive dates in con rmed cases that display a bar plot as a function of time in bottom panel (Figure 4 a-f). Display multiple plots in one gure in a mosaic type layout and generate a heatmaps comparing the daily con rmed cases changes per day in the different country (Figure 4 g).
The results showed that Japan, Philippines, India and Bangladesh had the highest number of daily con rmed cases on January 8, 2021(7863 people), August 4, 2020 (6263 people), September 10, 2020 (96511 people) and July 2 (4019 people), respectively (Figure 4 a, b, d e and, g). After that, the number of daily con rmed cases in three countries began to decline, the daily decline in the number of con rmed cases in Japan and Philippines has not been large and is therefore likely to rise again. However, the number of con rmed cases and the growth rate in a single day are still increasing in the Malaysia (con rmed cases, 5728/growth rate, 1.000524), and Indonesia (con rmed cases, 14518/growth rate, 1.051877), showing no sign of reaching the peak (Figure 4 c, f and g). , the number of con rmed cases increased by more than 13000 in a single day in Indonesia and India, which situation is still not encouraging.

Total con rmed cases, deaths per million people and CFR of the COVID-19 pandemic
The total con rmed cases, deaths per million people, CFR vs. tests per con rmed cases and tests per capita of the ongoing COVID-19 pandemic were calculated. Japan had the lowest and India had the highest in the number of con rmed COVID-19 cases per million India has the highest total number of cases and deaths and has the highest number of cases and deaths per million ( Figure S1 a, b) Malaysia has the lowest and Indonesia has the highest in CFR of the ongoing COVID-19 pandemic. However, Indonesia has the lowest and Malaysia has the highest in tests per con rmed case of the ongoing COVID-19 pandemic (Figure S1 c, d).

The trend of COVID-19 pandemic
COVID-19 pandemic were forecasted using Machine Learning method to project the spread of the SARS-CoV-2 based on the interventions and responses in the next 365 days by a basic model with no tweaking of season-related parameters and additional regression elements and prediction interval of 95%.
Under existing interventions and responses (such as social distance conditions) in each country ( Figure  1), Philippines, Bangladesh, India (on September, 2020) and Japan (on January, 2021) has peaked and The number of active cases are decreasing, however the number of active cases in the Malaysia and Indonesia is far from reaching its peak and is continuing to rise ( Figure 5 a-f).

Discussion
The interventions and the overall situation of the COVID-19 pandemic were evaluated by data-driven methods. India received low government response index scores in February, and the number of con rmed cases in September became quite high (Figure 1b, Figure 2 d), indicating that the early epidemic interventions and responses are very important. The median of Governmental response Stringency Index is 80.09 in Bangladesh with the most stringent interventions and responses (Figure 1a and b), India and the Philippines are also high, have shown a trend of slowdown in con rmed cases ( Figure 5) and decline in daily con rmed case (Figure 4). In Indonesia and Malaysia, The median of Governmental response Stringency Index is relatively low, and the number of total con rmed cases (Figure 2, 3, 5) and daily con rmed cases are on the rise so far (Figure 4 c and f). The number of active cases in Bangladesh, Japan and Philippines showed increasing trend. The number of active cases in India showed a trend of slowing down (168791, Figure 2d, Figure 5d) with large stock and the overall growth rate is higher than 1. The number of active cases in Malaysia and Indonesia showed a large stock and an increasing trend (47388 and 174083, Figure 2c and f, Figure 5 c and f) with the overall growth rate is always higher than 1 (Figure 3 c and f). The high number of active cases, growth rate well above 1 and the rising trend mean that the number of con rmed cases will continue to rise, and the situation will become more and more serious before the epidemic reaches its peak and develops for a long time.
Machine learning forecasts under current interventions and responses show that the Philippines, India, Japan and Bangladesh will be at and moderating, while Indonesia and Malaysia will still have no slowdown momentum and the situation will be severe ( Figure 5 a-f). The number of active cases in the Philippines, India, Bangladesh and Japan shows a downward trend after peaking in September, 2020 and January, 2021, respectively. The number of active cases in Malaysia and Indonesia is no sign of decline and the situation will be severe ( Figure 5 a-f, Figure 4 a-f).
These ndings show that although the total number of con rmed cases of COVID-19 is still increasing, the existing interventions and responses are still effective, because in general, the number of con rmed cases per million decreased with the increase of the government response stringency index. Interestingly, the higher the Tests per con rmed case is, the smaller the CFR will be. The number of daily con rmed cases and active cases of COVID-19 in Malaysia and Indonesia is no sign of decline, indicate the interventions and responses of these three countries still have room for improvement or do better.
Our study could have some limitations. First, the daily and total number of con rmed cases is not the actual daily number of infections and the number of total cases, because not all infected persons are tested, especially in countries that do not actively track, test and isolate suspected and close contacts, the number of con rmed cases and active cases may be much lower than true number of infections [21] [22] [23]. Only when a country or region can effectively isolate all infected and suspected persons, track, detect and isolate all close contacts, have strong detection capabilities and carry out large-scale screening, then the daily and total number of con rmed cases close to the true number of infections.
Second, our assessment and projection are based on the status quo that is the existing level of government Response, social distance, wearing masks and so on, but the status quo is unsustainable because all kinds of environment and conditions such as government response, social distancing, whether to wear masks are subject to change, and the country and society may need to gradually open up [5] [4], so the predicted results will also change accordingly. The peak of daily growth in some countries does not mean that they are in the clear, unless more stringent government measures, contact tracing and isolation, social distancing, and universal wearing of masks are maintained or even implemented.
Finally, the evaluation of the COVID-19 at this stage is very challenging based on the above reasons and the situation of the global epidemic, and the forecast of the cases numbers and trend of the COVID-19 must be carefully interpreted [24]. However, we are convinced that the results on cumulative case changes and machine learning predictions are much more reliable, because these are based on the shape of the COVID-19 epidemic curve to make machine learning predictions and not only based on the number of cases [17] [3] [15] [18]. We believe that widespread use of the vaccine could speed up the recovery, and our projections is useful to the effective interventions and taken against of the COVID-19 epidemic.

Conclusions
Data-driven method were used to assessment the status of the pandemic and machine learning method were used to forecast the COVID-19 pandemic in Asia-Paci c countries under the existing policy environment. This ndings show that although the total number of con rmed cases of COVID-19 is still increasing, the number of daily con rmed cases and active cases of COVID-19 in Japan, the Philippines, India and Bangladesh has begun to decline, while active cases and daily con rmed cases is no peak in Malaysia and Indonesia, indicate that the existing interventions and responses in Japan, the Philippines, India and Bangladesh are effective, while Malaysia and Indonesia need to be improved and strengthened. This study will be helpful in the response to COVID-19 pandemic.