Overview / Usage

As part of the Intel OneAPI hackathon, we have analysed COVID-19 trends in order to be better prepared for the future. The usecases are as follows:

1. Outbreak prediction based on disease spread rate

Using the datasets mentioned, the origin of it in the country and the R number for a microbe-generated disease like Covid, geography-based prediction of the spread of the disease has been predicted, which can help in formulating suitable plans for **implementing lockdown measures **by the Government.

1. Resource allocation

Using the spread rate predicted and the population density of various districts, the government can estimate the allocation and distribution of resources such as **food **and medicine. This prediction will help district authorities adequately prepare and equip people in each district to effectively navigate through the lockdown period.

1. Transportation route optimisation based on the google mobility report

By utilising the Google Mobility report and integrating it with the Google Maps API, we can suggest optimal transportation routes to citizens and government vehicles likewise. These routes can be recommended in order to avoid hotspots and ensure the efficient transportation and distribution of essential resources to different locations.

Due to budget constraints and limited project resources, we made the decision not to implement the Google Maps API in this project since it is a paid requirement. We are committed to delivering a high-quality project within the given constraints and have sought to maximize the available resources to provide meaningful insights and functionality to our users. But using Google Maps API could have proven to be a huge boost to our project.

1. Helping hospitals prepare & managing manufacturers' supply chain

Using the R number and the rates at which patients need treatment, an estimate of patient influx can be suggested to the hospitals by the governemnt. This will help them be better prepared with medical resources like hospital beds, masks etcetera.

With our solutions’ help, the Government can plan out appropriate ways to obtain resources from nearby manufacturing units and managing their supply chain. This can be done by obtaining a nation-wide data of **hospital bed manufacturers **and their weekly capacity.

1. Ensuring Government preparedness to tackle mental health & abuse case

Reports suggest a significant rise in cases of mental health issues, child abuse and domestic violence cases. With our solutions, once the Governemnt is aware of predicted threat of lockdown, the authorities can put adequate resources in place considering factors such as lockdown duration, AQI, financial strata and more of an area. In this way, we can help the Government be better prepared with the human resources required to tackle such issues. This includes special toll-free numbers for mental health experts, police personnel etcetera.

Methodology / Approach

Our methodology comprises a two-fold approach, leveraging technology to address the complex challenges posed by the COVID-19 pandemic:

​

1. Predictive Modeling with Transformers:

To forecast the potential number of future COVID-19 cases over specific time windows (e.g., one week or one month), we employ a state-of-the-art Transformers architecture. This choice is driven by the model's ability to capture long-term dependencies in time series data, which is crucial for mitigating noise and accounting for seasonality in the pandemic's progression. To enhance data quality, we apply a 7-day moving average to smoothen the time series data, which is then used as input sequences for the Transformer model.

The model is trained on 80 percent of the available data and evaluated on the remaining portion. We evaluate model performance using the Root Mean Square Error (RMSE) relative to the mean of the actual test data values, expressed as a percentage. The resulting error rate is approximately 1.25% for most counties, indicating a high level of predictive accuracy.

1. Risk Assessment and Visualization:

Our approach extends to making predictions for all counties in the United States for the next six months using the trained model. The predicted data is then visualized on an interactive map platform using Plotly. This map serves as a risk assessment tool, displaying the per capita probability of infection (risk) for each county.

The risk is calculated as the predicted number of cases in a county divided by the population of that county. This interactive map provides a valuable resource for both government authorities and the public. It aids in policy implementation by enabling informed decision-making and assists the public in taking appropriate precautions based on localized risk levels.

To gain deeper insights into the factors driving spikes in COVID-19 cases, we incorporate additional time series data sources such as mobility datasets, which capture changes in mobility across various areas like transit stations, workplaces, pharmacies, and grocery stores. We also include vaccination data. To identify correlations between these factors and case numbers, we employ the Kendall Tau correlation method. This analysis helps us understand the impact of different mobility factors on case spikes, empowering government bodies to implement targeted restrictions on activities that contribute to case surges while permitting those with minimal impact. This approach enhances our understanding of the pandemic's spread within specific regions, facilitating more effective containment strategies.

Technologies Used

Technologies

• Transformers
• Kendall Correlation
• Sklearn(Intel Patched Scikit Learn)
• Tensorflow(Intel TensorFlow Kit)
• Pandas(Modin)
• Plotly
• Scipy
• Numpy
• Scipy for Correlation analysis
• Plotly for plotting real time interactive maps and graphs

Intel technologies used in the development of this work.

Intel OneAPI AI Analytics Toolkit

Intel TensorFlow Kit

Repository

https://github.com/SparshRastogi/Intel-OneAPI

Other links

• Video