joker-testeur/neural-network-model-for-compartmenral-model-sir-and-seir-

Neural Network models for SIR and SEIR models

Guide: Dr. Adway mitra (Assistant Professor, Centre of Excellence in Artificial Intelligence, IIT Kharagpur)
Generated data of susceptible, infected, recovered cases using the SIR model for 100 days for 1000 different sets of parameters (beta and gamma) by using suitable initial conditions. Trained Neural network model using Keras to predict susceptible, infected, and recovered data for the next day using values of the previous day click here to see the notebook. Notice the changes when we did the same think but using parameters as one of the feature click here to see the notebook. Got validation accuracy of 0.98937 and test accuracy of 0.9880 for the model

In the next part I trained XGBoost Regression model to predict parameters (beta and gamma) separately one at a time click here to see the notebook. For beta prediction, got R2 score for training set as 0.99 and on the test as 0.91 and approximately similar results for gamma prediction. Extended the above tasks for the SEIR model a more complex form of the SIR model involving more parameters compared to the SIR model. Got approximately similar results as we got in case of SIR model.