Machine Learning For Combating COVID-19

The spread of coronavirus in over 200 countries within a span of a few months necessitates the development of forecasting models that can assist governments around the globe in keeping an eye on their country’s situation forecast so that preventive measures can be timely taken. Since the epidemic spread is still in the early stages in some countries, forecasting models harnessing data from countries already hit by the epidemic will provide a realistic and timely intimation of risks of potential outbreaks in these countries. The proposed open-source data and machine learning model will provide useful insights into the disease spread spectrum, diagnostics, and prevention strategies. Data scientists around the world will be encouraged to pool in their efforts and augment our model to assist the governmental entities in taking emergency response measures ahead of time during a pandemic outbreak.

We have developed several machine learning models for making predictions on various real-world problems. Our lab has recently completed a project (Sep 2017 - Aug 2019) funded by the Pakistan Science Foundation titled “Artificial Intelligence-based Diagnostic Screening for Dengue Fever”. Since the forecasting and analysis of COVID-19 require the same domain expertise with fundamental knowledge of Machine Learning and simulation models, the skills of the technical team that worked on the aforementioned project and the algorithms developed are being utilized again. Data is being gathered from several sources like publicly available pandemic spread data, data from local hospitals and health institutions. Deep learning based forecasting and prediction algorithms are being trained on the collected data.

The project is currently being run using volunteer time dedicated by different students, researchers, and postdocs under the leadership of Prof. Dr. Faisal Shafait. Stay tuned for data driven insights about COVID-19!

Artificial Intelligence Based Insights about COVID-19

Figure 1 show the results of our forecasting model on the number of new cases in four countries (Italy, Germany, China, and Australia). When we use this data to do forecast for Pakistan – a country where epidemic spread started late and hence the data from countries affected earlier can be used in the forecasting model. Our forecasting model is based on a custom architecture of Deep Long Short-Term Memory (LSTM) Neural Networks and is capable of using transfer learning to adapt the model based on limited availability of real-world data.

Figure 1: Forecasting next seven days of new cases in countries affected for longer period of time by COVID-19 as compared to Pakistan. Our model is showing a steady trend for almost all the countries, which indicates that a state of maturity might have been reached and better preventive measure have been implemented to restrict the sharp increase in the cases.

Figure 2:Forecast of COVID-19 spread in Pakistan using deep learning. The blue line shows the daily number of COVID-19 cases reported in Pakistan per day, whereas the orange line shows the prediction results.

Team Members

Prof. Dr. Faisal Shafait
Professor, NUST SEECS
1. Dr. Adnan Ul-Hasan
  Post-Doctorate
2. Maham Jahangir
  PhD Student
3. Nosheen Abid
  PhD Student
4. Ahsan Jalal
  PhD Student
5. Mahnoor Haneef
  Masters Student
1. Abdul Wahab
  Bachelor Student
2. Muhammad Kamran Janjua
  Bachelor Student
3. Raja Hasnain Anwar
  Bachelor Student
4. Saifullah
  Bachelor Student