1680
Time Series Analysis For Predicting Covid-19 Infection Using Facebook Prophet Model
1Dr.A.Bazila Banu*, 2Dr.P.Thirumalaikolundusubramanian
1Professor,Department of Computer Science Engineering, Bannari Amman Institute of Technology, Erode, India
(Affiliated to Anna University, Chennai).
2Departments of Microbiology, General Medicine1 and Community Health Services, Trichy SRM Medical
College Hospital and Research Centre, Tiruchirapalli, India (Affiliated to the Tamilnadu Dr. MGR Medical University, Chennai),
1bazilabanu@bitsathy.ac.in 2 ponniah.tks@gmail.com
*Corresponding Author E-mail: bazilabanu@bitsathy.ac.in
Article History: Received: 11 January 2021; Revised: 12 February 2021; Accepted: 27 March 2021; Published
online: 16 April 2021
ABSTRACT: The International fright due to the occurrence of COVID has created an emergency act in the field of
healthcare, bio medical and drug discovery process. However, finding the feasible solution to introduce a drug is a time-consuming process due to pre-clinical and post-clinical testing process. Prediction and estimation of COVID-19 can help the medical practitioners and government authorities to take preventable measure against the outcomes of COVID-19.From December 2019 to April 2020, 2844712 cases of COVID-19 have been informed, which includes 201315 deaths according to European Centre for Disease Prevention and Control. This drastic condition should be treated not only physicians and other health care providers. There are two types of time series forecasting techniques. The first technique time-domain approach models the forthcoming values as a function of previous and current values. The groundwork of this approach is the time series regression of current values of a time series on its own past values. The assessments of the model are applied for forecasting process. The second technique known as Frequency domain models are based on the interpretations of time using sines and cosines functions. These interpretations are known as Fourier representations. Overall, the technique utilizes regressions on sines and cosines function, to model the behavior of the data. The proposed work used Facebook Prophet model for Time Series Analysis to forecast the trend for the year 2021. The models will act as an inference tool to take decisions during pandemic conditions
Keywords: Time Series Analysis(TSA) ,Corona Virus Disease-19(COVID-19),Change Point Detection (CPD),Facebooks
Prophet(FB Prophet), Artificial Neural Network(ANN)
INTRODUCTION:
The statistical attributes of time series data often contravene the expectations of traditional statistical methods. As a result, analyzing time series data needs a unique set of tools and methods, collectively known as Time Series Analysis(TSA).Moreover TSA requires more time to accomplish data preprocessing in order to fit the data for training since it is a collection of quantities that are accumulated over even intervals in time and well-organized chronologically. A time series graph plots the attributes time and observer values on x-axis and y-axis . These graphs will act as an inference tool to identify the patterns present in the trend .Visualizing such graphs adds more clarity to make quick decisions .This has become true especially in the Corona Virus Disease -19 (COVID-19) prediction and analysis .Day by day the spread of virus needs to be monitored in order to implement the remedial protocol measure known as lockdown which prevents the people leaving a given area. Government has taken various kind of methods to announce the lock down procedure .However detailed investigation about time series analysis of COVID-19 may help to decide the lock down protocol in a better manner ,related to particular zone .TSA is used to identify trend, seasonality and structural breaks present in the graph.
TSA methods like Naïve ,Exponential Smoothing ,Auto Regressive Moving Average (ARIMA) ,Multi-Layer Perceptron, Recurrent Neural Network ,Long Short Term Memory(LSTM) are available .However in spite of producing good accuracy these models needs to accomplish huge amount of time in preprocessing the data like transforming the data from one form to another form in order to apply the method .Hence the proposed work uses Facebooks Prophet(FB Prophet) method to predict the future trend as well as to detect the changepoints known as Change Point Detection (CPD) .Its a method to recognize the sudden change in data when a characteristic of time-series changes .
Prophet is open source software released by Facebook’s Core Data Science
team(https://facebook.github.io/prophet/). It is available for download on CRAN and PyPI. FB Prophet is a open source method for forecasting time series data based on an additive model to forecast non-linear trends with attributes like yearly, weekly, daily effects of trends. Prophet is robust to misplaced data and swings in the trend, and naturally handles outliers well. It performs best with cyclical effects and numerous seasons of historical data
Time Series Analysis For Predicting Covid-19 Infection Using Facebook Prophet Model
1681 and reduces the time for preprocessing and data transformation approaches. This is turn improves the computational speed while training the data to predict the future trend.
RELATED WORKS:
Product-level sales forecasts are an imperative factor in the retail business since stock control and creation arranging assumes a significant function in the seriousness of any organization that gives merchandise to its clients. While precise and solid gauges can prompt immense reserve funds and cost decreases by encouraging better creation and stock arranging, serious valuing and convenient advancement arranging, helpless deals assessments are demonstrated to be expensive in this area since it is notable that products deficiencies cause lower benefits and can effectively lead to client disappointment. Moreover, not just the abundance stock may drive the store to sell merchandise at lower costs, or surprisingly more dreadful lead to stock benefits, higher than required stock levels additionally increment warehousing costs. In reality situation, the business climate in the retail business is exceptionally unique and regularly unstable, which is transcendently brought about by occasion impacts and contender conduct [10]. As a result, in opposition to the generally accessible scholarly datasets used to exhibit and benchmark different time-arrangement guaging techniques, certifiable deals information in this area convey different challenges, for example, profoundly non-fixed recorded information, sporadic deals designs, and exceptionally discontinuous deals information. A module that would have the option to figure deals with a sensibly high exactness, increased by the module for profoundly dependable order of the item portfolio as indicated by the normal level of forecasting, would be of incredible use for any organization working in the retail business [2].
The fundamental structure of the system for deals determining and item portfolio characterization is an apparatus/strategy for creating top notch time-arrangement gauges. Regardless of the reality that there are various devices/strategies that can be applied, it was chosen to utilize Facebook's Prophet device for this exploration since it is fit for creating conjectures of a sensible quality at scale. As indicated by Prophet is utilized in numerous applications across Facebook for creating solid forecasting and performs in a way that is better than some other methodology in most of cases [11]. Facebook's Prophet device is utilized for displaying the elements of deals for things in a item portfolio without utilizing extra regressors, with the point of producing month to month and quarterly deals estimates. It merits referencing that an exact strategy for tweaking model boundaries is utilized to join space information into the proposed structure; however, the equivalent boundaries are utilized for the whole item portfolio to abstain from overfitting. It is observationally inferred that in any event two years of authentic information is needed for dependable assessment of pattern and additionally occasional impacts [9].
Forecasting is a typical data science task that helps associations with objective setting, and peculiarity recognition. In spite of its significance, there are genuine difficulties related with delivering dependable and great conjectures – particularly when there is an assortment of time arrangement and investigators with ability in time arrangement demonstrating are generally uncommon. To address these difficulties, portray a commonsense way to deal with anticipating "at scale" that consolidates configurable models with examiner on the up and up execution investigation [9]. A relapse model with interpretable boundaries that can be naturally changed by investigators with area information about the time arrangement. Tools that help examiners to utilize their mastery most adequately empower solid, commonsense estimating of business time arrangement There is a wide variety of business forecasting issues; anyway, there are a few highlights normal to huge numbers of them. Facebook clients can utilize the Events stage to make pages for functions, welcome others, and collaborate with functions in an assortment of ways. There are a few occasional impacts plainly noticeable in this time arrangement: week by week and yearly cycles [8].
Prophet is an open-source device from Facebook utilized for determining time arrangement information which assists organizations with comprehension and conceivably foresees the market. It depends on a decomposable added substance model where non-straight patterns are fit with irregularity. it likewise considers the impacts of occasions. The pattern shows the inclination of the information to increment or lessening over a significant stretch of time and it sift through the occasional varieties. Irregularity is the varieties that happen over a brief timeframe is not noticeable enough to be known as a "pattern" [4]
Time series investigation is a way to deal with examines time arrangement information to extricate significant attributes of information and produce other helpful bits of knowledge applied in business circumstance. By and large, time-arrangement information is a succession of perceptions put away in time request. Time-arrangement information frequently stands apart when following business measurements, checking modern cycles and so on. Time arrangement investigation comprehends time-based examples of a lot of metric information focuses which is basic for any business. The essential goal of time arrangement examination ordinarily is to decide a model that portrays the example of the time arrangement and could be utilized for anticipating. Old style time
1682
arrangement determining methods expand on details models which requires bunches of exertion to tune models and expect in information and industry. The individual needs to tune the boundaries of the technique concerning the issue when an anticipating model does not proceed true to form. Tuning these techniques requires an exhaustive comprehension of how the basic time arrangement models work. It's hard for certain associations to taking care of those anticipating without information science groups. Also, it may appear to be doesn't beneficial for an association to have a lot of expects ready if there is no need a form a mind boggling determining stage or different administrations [5].
The continuous COVID-19 pandemic has caused overall financial turmoil, driving governments to acquaint extraordinary measures with lessen its spread [1]. Having the option to precisely conjecture at the point when the flare-up will hit its pinnacle would essentially reduce the effect of the illness, as it would permit governments to change their approach in like manner and plan ahead for the preventive advances required, for example, general wellbeing informing, bringing issues to light of residents and expanding the limit of the wellbeing framework [7]. This examination explored the exactness of an assortment of time arrangement demonstrating approaches for COVID-19 episode location in ten distinct nations with the most noteworthy number of affirmed cases starting at 4 May 2020. For every one of these nations, six distinctive time arrangement draws near were created and looked at utilizing two freely accessible datasets with respect to the movement of the infection in every nation and the number of inhabitants in every nation, separately. The outcomes exhibit that, given information created utilizing real testing for a little bit of the populace, AI time arrangement strategies can learn and scale to precisely assess the level of the all-out populace that will get influenced later on [6].
Facebook delivered a bundle actualizing a Bayesian forecasting approach. This strategy perceives rehashing designs over weeks, months, a long time, and distinguished occasions. Prophet is set up as a mechanized cycle and can be introduced as a bundle in R or Python. The fundamental approach is an iterative bend coordinating schedule, where Prophet will at that point train your information on a greater period, at that point anticipate again and this will rehash until the end point is reached. It is Working with high-recurrence information (hourly, every day, or week by week) with numerous seasonality, for example, hour of day, day of week and season. Exceptional functions and bank occasions are not fixed in the year. By taking into consideration the presence of a sensible number of missing qualities or huge anomalies to represents the changes in the chronicled patterns and non-straight development bends in a dataset. Further preferences incorporate the capacity to prepare from a moderate estimated dataset, without the requirement for authority business programming, and quick beginning up times for development5. While further developed models are created, time-arrangement based [12]. Table.1 illustrates the comparison study about the forecasting methods like SARIMA,Exponential smoothing,ARIMA, Prophet,TBATS and Artificial Neural Network(ANN).
Fore casting Methods
Advantage Disadvantage
SARIMA It is used for modeling seasonal variation, trend and it does not require any external data
It is less accurate than linear regression
Exponential smoothing
Fast and fully automated method for modelling seasonal variation, trends.
It is less accurate while analyzing the cyclical data ARIMA It is widely used for modeling and
provide interpretable results
It requires complex statistical expression to analyze the data in an effective manner
Prophet It uses simple syntax and fully
automated environment for modelling
Computational time is increased while tuning
TBATS It is dynamic and automated process Computational time is
increased while modeling the large dataset
ANN It is adjusted for series time change and handle complex systems
Complex to train the model and it is not transparent Table.1 :Comparison of Forecasting Methods
Among the forecasting methods Prophet can be used for systems that need to be automated .
Time Series Analysis For Predicting Covid-19 Infection Using Facebook Prophet Model
1683 All the data gathered is made freely available, primarily as google sheets, in a GitHub repository(https://codeload.github.com/CSSEGISandData/COVID-19). Data pertaining to India was extracted from 1st March to 30th September 2019 for forecasting process. Parameters like date and the total number of cases recorded per day are used for the study. For making the analysis Facebook Prophet Model (Taylor and Letham, 2018) and the packages from python has been accessed. Prophet package contains the methods for constructing the model by substituting the time series data such as day and the number of cases .Packages like plot_plotly, plot_components_plotly were used to forecast the data .
Prophet supports the Sklearn package available in Python. An instance of the Prophet class and to invoke the fit and predict methods are implemented for COVID-19 prediction. The input to Prophet consist of dataframe with two attributes such as ‘date stamp’ and ‘count’ . The first attribute ds should be like YYYY-MM-DD for a date or YYYY-MM-DD HH:MM:SS for a timestamp. The second attribute ‘count’ should be numeric for the measurement value to be forecasted. As part of preprocessing the data the format of the date stamp present in the dataset has been transformed into the required format using the datetime conversion method available in Pandas package.
There are two types of time series forecasting techniques. The first technique time-domain(TD) approach models the forthcoming values as a function of previous and current values. The groundwork of this approach is the time series regression of current values of a time series on its own past values. The assessments of the model are applied for forecasting process. The second technique known as Frequency Domain(FD) models are based on the interpretations of time using sines and cosines functions. These interpretations are known as Fourier representations. Overall, the technique utilizes regressions on sines and cosines function, to model the behavior of the data.
Among the two techniques time-domain approach model is used to forecast the COVID-19 infection for the forthcoming days/months and year. The model is constructed by creating Prophet object and the method known as fit is invoked by passing the attributed date stamp and count available in the dataframe .Predictions are then made on a dataframe with a attribute date stamp comprising the dates. The helper method present in prophet package Prophet.make_future_dataframe was used to predict the future COVID-19 cases .
RESULTS AND DISCUSSION
Prophet utilizes a linear model for its forecast. While forecasting COVID-19 growth , the results shows that the total number of cases are increasing in a linear fashion .The results are plotted in a graph for predicting the trend on daily/monthly (Figure 1) and yearly basis (Figure 2).
1684
Fig.2:COVID-19 Prediction for 2021 Report
When forecasting growth, there is usually potential number of changepoints could be inferred which helps us to understand the real trend .Changepoint convey the rate of change over the period .Prophet distinguishes changepoints by specifying a large number of probable changepoints at which the rate is permitted to change. The changepoint results are observed for the data(Figure 3) by providing the changepoint value to be minimal as 0.005. When visualizing the forecast, this parameter can be changed as required if the trend appears to be over- or under-fit. Based on the results shown in the Figure 3 the change point has been observed for September 2020 .
Fig. 3:COVID-19 Change Point Detection CONCLUSION
The Analysis of Indian data for COVID-19 using Facebook Prophet model for Time Series Analysis to forecast the trend for the year 2021 shows that there will be drastic increase in the number of cases after three months starting from April 2020 onwards Time-domain approach model to forecast the COVID-19 infection for the forthcoming days/months and year clearly shows the analysis in an linear fashion .However after applying the
Time Series Analysis For Predicting Covid-19 Infection Using Facebook Prophet Model
1685 change point deduction method it is observed that the number of cases will reduce from September to October 2020 and then increase considerably. Overall, the prophet model is used to analyze the complete trend COVID-19 data and sufficient measures need to be taken to overcome this scenario.
CONFLICT OF INTEREST:
The authors declare no conflict of interest.
REFERENCES
1. ANG, Y., NIE, Y. & PENNY, M. 2020. Transmission dynamics of the COVID‐19 outbreak and effectiveness of government interventions: A data‐driven analysis. Journal of medical virology, 92, 645-659.
2. Bangladesh COVID-19 Daily Cases Time Series Analysis using Facebook Prophet Model. Available
from:
https://www.researchgate.net/publication/343306716_Bangladesh_COVID-19_Daily_Cases_Time_Series_Analysis_using _Facebook_Prophet_Model [accessed Oct 26 2020]. 3. Emir Žunić1,2, Kemal Korjenić1 , Kerim Hodžić2,1 and Dženana Đonko2, Application of Facebook's
Prophet Algorithm For Successful Sales Forecasting Based on Real-World Data, International Journal of Computer Science & Information Technology (IJCSIT) Vol 12, No 2, April (2020).
4. Sean J. Taylor, Benjamin Letham, Forecasting at Scale, Preprints |
https://doi.org/10.7287/peerj.preprints.3190v2 | CC BY 4.0 Open Access | rec: 27, publ: 27 Sep (2017). 5. Tholkapiyan, A.Mohan, Vijayan.D.S, A survey of recent studies on chlorophyll variation in Indian
coastal waters, IOP Conf. Series: Materials Science and Engineering 993 (2020) 012041, 1-6.
6. https://towardsdatascience.com/a-quick-start-of-time-series-forecasting-with-a-practical-example-using-fb-prophet-31c4447a2274
7. Vasilis Papastefanopoulos * , Pantelis Linardatos and Sotiris Kotsiantis, COVID-19: A Comparison of Time Series Methods to Forecast Percentage of Active Cases per Population (2020), https://www.mdpi.com/journal/applsci
8. Al-qaness, M.A et.all M. Optimization method for forecasting confirmed cases of covid-19 in China. J. Clin. Med. (2020)
9. World Health Organization. Naming the Coronavirus Disease (COVID-19)
https://www.who.int/emergencies/diseases/novelcoronavirus-2019/technical-guidance/naming-the-coronavirus-disease-(covid-2019)-and-the-virus-thatcauses-it (2020).
10. Rajaram, A., & Lingam, S. V. (2011). Distributed Adaptive Clustering Algorithm for Improving Data Accessibility in MANET. International Journal of Computer Science Issues (IJCSI), 8(4), 369.
11. Gopalakrishnan, R., Mohan, A., Sankar, L. P., & Vijayan, D. S. (2020). Characterisation On Toughness Property Of Self-Compacting Fibre Reinforced Concrete. In Journal of Environmental Protection and Ecology (Vol. 21, Issue 6, pp. 2153–2163).
12. S Aras, İ Deveci Kocakoç, C Polat, Comparative study on retail sales forecasting between single and combination methods, (2017)
13. I Alon, M Qi, R J Sadowski, Forecasting aggregate retail sales: a comparison of artificial neural networks and traditional method, Journal of Retailing and Consumer Services,(2001)
14. Advanced forecasting techniques, https://www.england.nhs.uk/wp-content/uploads/(2020)/01/advanced-forecasting-techniques.pdf
