• UVA models:
  • AR_analogues and ARIMA: This class of linear methods model the signal to be forecast using its lagged versions. We also incorporate the population-normalized lagged time series data of other states and countries. Since the time series under consideration is typically non-stationary, we log-transform it (to nullify the large variations in variance across time) and train the model every week over short segments with the assumption that the signal is relatively stationary over that period. We use a subset selection method to obtain a sparse set of predictors. The more general non-seasonal Autoregressive Integrated Moving Average (ARIMA) are effective for modeling signals with some degree of non-stationarity (trends), by specifying three parameters: autoregressive lag parameter p, the order of differencing d, and the order of the moving average filter q. We employ the popular forecast package to determine the parameters.
  • LSTM: This is a deep-learning model and uses the Long Short-Term Memory (LSTM) networks to capture the temporal dynamics of COVID-19 time series. The model was implemented as one LSTM layer with hidden layer of size 32, one dense layer with hidden layer of size 16, a rectified linear unit activation function, and one dropout layer (dropout rate of 0.2). The output layer is a dense layer with linear activation and L2 kernel regularization (0.01 penalty factor). The historical window size is 3 weeks. We use the mean squared error (MSE) loss function and train the model with the Adam optimizer with a batch size of 32. Probabilistic forecasts are generated using MCDropout. In order to avoid overfitting, we train a single model using time series data from all the states.
  • SEIR-adpt: This method uses an Susceptible-Exposed-Infectious-Recovered (SEIR) model and train each state in isolation, where effects of social distancing and miscellaneous adaptations are captured as temporal variations in the model's transmissibility parameter. Using a simulation optimization approach, we sequentially estimate the time-varying transmissibility parameter with appropriate delays and scaling applied from simulated infections to confirmed cases. For each t, the estimation is done using Golden Section Search (GSS), a maximally efficient extremum search method within a specified interval. Each state’s confirmed cases time series is fit precisely through a daily varying transmissibility, and a smoothed version of recent transmissibility is used for projections/forecasting.
  Contributors: Aniruddha Adiga, Lijing Wang, Benjamin Hurt, Akhil Peddireddy, Przemyslaw Porebski, Srinivasan Venkatramanan, Bryan Lewis, Madhav Marathe
  
• IISc-ISI models:
  • SIR-IISc-ISI: This prediction is based on a SIR model. In this model, the entire population is categorized into three states- Susceptible, which refers to healthy individuals at a risk of contracting the disease; Infected, which refers to the people infected with the disease; and Recovered, which refers to people who have already contracted and then recovered from the disease. At any time instant, an individual of the population falls in one and only one of the above three categories. For a detailed description of fit, please click here.
  • Log Linear-IISc-ISI: Here we consider log of number of cases on each day. We use an automated window to find the best least-square fit as described in IISC-ISI Prediction 1. We also use the same procedure for providing the error band for the one week future prediction.
  • Omicron IISc-ISI: This is a compartmental SEIR model to capture the resurgence due to the Omicron variant. The key parameter in this model is the contact rate $\beta$ that models the transmissibility of the virus. We first calibrate this model on the reported cases from South Africa during the Delta wave (15 May - 15 June 2021) and the Omicron wave (15 November - 12 December 2021) to compute the contact rates for the Delta and Omicron variants. We calculate the ratio of the Omicron's calibrated contact rate to that of the Delta's. This is the transmission advantage of Omicron compared to Delta, which we assume is applicable to the Indian population as well. We then apply this factor to the calibrated contact rates for the Indian states during the Delta wave (15 March - 30 April 2021) and run the projections forward for the Omicron variant. We assume that 60% of the population is susceptible to the Omicron variant. More details can be found in the attached PDF.
  Contributors:Siva Athreya, Nihesh Rathod, A.Y. Sarath, Rajesh Sundaresan
  
• SUTRA* model:
  • Implementation of SUTRA model by IISc team.
• CSIR-4PI LSTM model:
  • Advanced and accurate forecasting of COVID‐19 cases plays a crucial role in planning and supplying resources effectively. Artificial Intelligence (AI) techniques have proved their capability in time series forecasting non‐linear problems. In the present study, the relationship between weather factor and COVID‐19 cases was assessed, and also developed a forecasting model using long short‐term memory (LSTM), a deep learning model. The study found that the specific humidity has a strong positive correlation, whereas there is a negative correlation with maximum temperature, and a positive correlation with minimum temperature was observed in various geographic locations of India. The weather data and COVID‐19 confirmed case data (1 April to 30 June 2020) were used to optimize univariate and multivariate LSTM time series forecast models. The optimized models were utilized to forecast the daily COVID‐19 cases for the period 1 July 2020 to 31 July 2020 with 1 to 14 days of lead time. The results showed that the univariate LSTM model was reasonably good for the short‐term (1 day lead) forecast of COVID‐19 cases (relative error <20%). Moreover, the multivariate LSTM model improved the medium‐range forecast skill (1–7 days lead) after including the weather factors. The study observed that the specific humidity played a crucial role in improving the forecast skill majorly in the West and northwest region of India. Similarly, the temperature played a significant role in model enhancement in the Southern and Eastern regions of India.
  Contributors:Gopal Krishna Patra, Kantha Rao Bhimala, Rajasekhar Mopuri and Srinivasa Rao Mutheneni
  
• IISc-CoviHawkes model:
  • Hawkes-LSTM: The observed patterns in the case counts from the past with additional factors like demographics and mobility of the region are combined in order to make predictions. An LSTM is used to model the reproduction number R(t) of the Covid-19 virus using past mobility data. A higher value of R(t) entails faster spread of the disease and vice versa. Due to the emergence of multiple variants of the virus R(t) changes over time and is estimated periodically using recent trends in the data. To validate the short term predictions, case counts and mobility features collected between March 2, 2020, and July 20, 2021 are used. This data is divided into two subsets: training data (March 2, 2020 to April 27, 2021) and validation data (April 27-July 20, 2021), and the model is validated for forecasting windows of three sizes: 7, 14, and 28 days. More details can be found in the attached PDF.
  Contributors:Ambedkar Dukkipati, Tony Gracious, Shubham Gupta, Abhishek Singh Narwaria
  

This repository has been developed to provide a common platform for modelling and forecasting teams to contribute short-term COVID-19 incident case forecasts. Our goal is to enable effective communication of available forecasts to both the public and the policy makers. Using multiple model forecasts, we provide ensemble model forecasts, which have been shown to be robust and accurate [1]. Motivated by similar efforts in the United States of America (https://covid19forecasthub.org/) and the European Union (https://covid19forecasthub.eu/), which consist of many working groups of academic and industry-based forecasters, this open-science platform provides standardized robust forecasts, ground truth data, evaluation methods, and visualization tools. These efforts are built based on more than a decade of multi-model efforts in seasonal influenza forecasting in the United States [1]. We believe that with this common platform and standardized tools, contributing members can collectively evaluate and refine their models, thus leading to a better understanding of the contagion process. In addition, the development of ensemble models is supported by ongoing research efforts [2] This collaborative effort is being coordinated by a team comprising experts in infectious disease modeling, statistics, information theory, network theory and machine learning, with experience in advising public health agencies in India [3,4,5] and the United States of America [6,7]. Several of the team members have been contributing forecasts for over a year to the COVID-19 Forecast Hub and the EU Forecast Hub. In addition to COVID-19, the team has been participating for over 10 years in modeling and collaborative forecasting of influenza-like illness [8,9] and Dengue.

We thank the COVID-19 Forecast Hub and the EU Forecast Hub for providing a template for collaborative forecasting. This effort is funded by NSF RAPID: Modeling and Analytics for COVID-19 Outbreak Response in India: A multi-institutional, US-India joint collaborative effort (Award Number (FAIN): 2142997), NSF Expeditions in Computing Grant CCF-1918656, and in part by an IOE project grant from the Indian Institute of Science and support in part by the Centre for Networked Intelligence, Indian Institute of Science and Indian Statistical Institute.

We source the data from https://www.mohfw.gov.in/.
Unprocessed: [Daily India States Data in CSV].
Processed: [Daily India States Data in CSV], and [Weekly India States Data in CSV]

1. Reich, Nicholas G., et al. A collaborative multiyear, multimodel assessment of seasonal influenza forecasting in the United States. Proceedings of the National Academy of Sciences 116.8 (2019): 3146-3154.
2. Reich, Nicholas G., et al. O the predictability of COVID-19
3. Athreya, S., Babu, G.R., Iyer, A., Rathod, N., Shriram, S., Sundaresan, R., Vaidhiyan, N.K. and Yasodharan, S., 2020. COVID-19: Optimal Design of Serosurveys for Disease Burden Estimation. arXiv preprint arXiv:2012.12135.
4. Adiga, A., Athreya, S., Lewis, B., Marathe, M.V., Rathod, N., Sundaresan, R., Swarup, S., Venkatramanan, S. and Yasodharan, S., 2021. Strategies to Mitigate COVID-19 Resurgence Assuming Immunity Waning: A Study for Karnataka, India. medRxiv.
5. Talekar, A., Shriram, S., Vaidhiyan, N., Aggarwal, G., Chen, J., Venkatramanan, S., Wang, L., Adiga, A., Sadilek, A., Tendulkar, A. and Marathe, M., 2020. Cohorting to isolate asymptomatic spreaders: An agent-based simulation study on the Mumbai Suburban Railway. Extended Abstract at AAMAS 2021.
6. Adiga, A., Wang, L., Hurt, B., Peddireddy, A.S., Porebski, P., Venkatramanan, S., Lewis, B. and Marathe, M., 2021. All models are useful: Bayesian ensembling for robust high resolution covid-19 forecasting. ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2021.
7. Borchering, R.K., Viboud, C., Howerton, E., Smith, C.P., Truelove, S., Runge, M.C., Reich, N.G., Contamin, L., Levander, J., Salerno, J. and van Panhuis, W., 2021. Modeling of future COVID-19 cases, hospitalizations, and deaths, by vaccination rates and nonpharmaceutical intervention scenarios--United States, April--September 2021. Morbidity and Mortality Weekly Report, 70(19), p.719.
8. Biggerstaff, M., Alper, D., Dredze, M., Fox, S., Fung, I.C.H., Hickmann, K.S., Lewis, B., Rosenfeld, R., Shaman, J., Tsou, M.H. and Velardi, P., 2016. Results from the centers for disease control and prevention's predict the 2013-2014 Influenza Season Challenge. BMC infectious diseases, 16(1), pp.1-10.
9. Halloran, M.E., Ferguson, N.M., Eubank, S., Longini, I.M., Cummings, D.A., Lewis, B., Xu, S., Fraser, C., Vullikanti, A., Germann, T.C. and Wagener, D., 2008. Modeling targeted layered containment of an influenza pandemic in the United States. Proceedings of the National Academy of Sciences, 105(12), pp.4639-4644.